Deploy Second Life Flask app

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+Architecture[[:space:]]Diagrams/secondlife[[:space:]]high[[:space:]]level[[:space:]]architecture.png filter=lfs diff=lfs merge=lfs -text

.hfignore

@@ -0,0 +1,16 @@
+.git/
+.claude/
+__pycache__/
+.venv/
+.python_packages/
+.tmp/
+
+Final Clinical Trails Data/
+Final Patients Synthea Data/
+mimic-iv-clinical-database-demo-2.2/
+*.zip
+
+server.log
+*.log
+*.tmp
+UI/

Architecture Diagrams/secondlife high level architecture.drawio

@@ -0,0 +1,163 @@
+<mxfile host="app.diagrams.net">
+  <diagram id="E15GttQApXdkw6PukwBz" name="Page-1">
+    <mxGraphModel dx="2322" dy="2022" grid="1" gridSize="10" guides="1" tooltips="1" connect="1" arrows="1" fold="1" page="0" pageScale="1" pageWidth="1600" pageHeight="900" math="0" shadow="0">
+      <root>
+        <mxCell id="0" />
+        <mxCell id="1" parent="0" />
+        <mxCell id="userGroup" parent="1" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#EEF4FF;strokeColor=#C7D8FF;strokeWidth=3;arcSize=18;opacity=70;fontSize=12;" value="" vertex="1">
+          <mxGeometry height="356.6470588235295" width="1824.563758389262" x="142.5503355704698" y="18.999411764705883" as="geometry" />
+        </mxCell>
+        <mxCell id="bg1" parent="1" style="ellipse;whiteSpace=wrap;html=1;fillColor=#DCEBFF;strokeColor=none;opacity=55;" value="" vertex="1">
+          <mxGeometry height="394.59" width="383.76" x="1676.24" y="-110" as="geometry" />
+        </mxCell>
+        <mxCell id="bg2" parent="1" style="ellipse;whiteSpace=wrap;html=1;fillColor=#DFF7EC;strokeColor=none;opacity=60;" value="" vertex="1">
+          <mxGeometry height="364.24" width="367.32" y="815.76" as="geometry" />
+        </mxCell>
+        <mxCell id="bg3" parent="1" style="ellipse;whiteSpace=wrap;html=1;fillColor=#FFF1D7;strokeColor=none;opacity=65;" value="" vertex="1">
+          <mxGeometry height="273.1764705882353" width="237.98657718120808" x="1464.6979865771812" y="815.7647058823532" as="geometry" />
+        </mxCell>
+        <mxCell id="title" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=40;fontStyle=1;fontColor=#152641;align=center;verticalAlign=middle;" value="Second Life - High-Level Architecture" vertex="1">
+          <mxGeometry height="60.70588235294119" width="1216.3758389261745" x="446.64429530201346" y="-106.96470588235294" as="geometry" />
+        </mxCell>
+        <mxCell id="sec1" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=18;fontStyle=1;fontColor=#61728E;letterSpacing=1;" value="USER EXPERIENCE" vertex="1">
+          <mxGeometry height="36.423529411764704" width="224.76510067114094" x="261.5436241610738" y="49.35294117647058" as="geometry" />
+        </mxCell>
+        <mxCell id="sec3" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=18;fontStyle=1;fontColor=#61728E;letterSpacing=1;" value="DATA AND INTELLIGENCE" vertex="1">
+          <mxGeometry height="36.423529411764704" width="290.8724832214765" x="119.99624161073825" y="703.4629411764703" as="geometry" />
+        </mxCell>
+        <mxCell id="patientCard" parent="1" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#EEF4FF;strokeColor=#5C8CF0;strokeWidth=3;arcSize=18;shadow=1;" value="" vertex="1">
+          <mxGeometry height="197.29411764705884" width="661.0738255033558" x="248.32214765100673" y="87.29411764705884" as="geometry" />
+        </mxCell>
+        <mxCell id="patientBadge" parent="1" style="ellipse;whiteSpace=wrap;html=1;fillColor=#5C8CF0;strokeColor=#5C8CF0;fontColor=#FFFFFF;fontSize=18;fontStyle=1;align=center;verticalAlign=middle;" value="P" vertex="1">
+          <mxGeometry height="72.85" width="74.67" x="290" y="132.82" as="geometry" />
+        </mxCell>
+        <mxCell id="patientTitle" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=28;fontStyle=1;fontColor=#152641;align=left;" value="Patient Portal" vertex="1">
+          <mxGeometry height="51.60000000000001" width="343.758389261745" x="387.1476510067114" y="144.96470588235297" as="geometry" />
+        </mxCell>
+        <mxCell id="patientBody" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=20;fontColor=#5E708C;align=left;verticalAlign=top;" value="Profile, suggested trials, saved interests&lt;br&gt;Connections, inbox chat, document uploads" vertex="1">
+          <mxGeometry height="69.81176470588235" width="436.30872483221475" x="387.1476510067114" y="205.67058823529413" as="geometry" />
+        </mxCell>
+        <mxCell id="hospitalCard" parent="1" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#EEF8F2;strokeColor=#41A56C;strokeWidth=3;arcSize=18;shadow=1;" value="" vertex="1">
+          <mxGeometry height="197.29411764705884" width="661.0738255033558" x="1200.268456375839" y="87.29411764705884" as="geometry" />
+        </mxCell>
+        <mxCell id="hospitalBadge" parent="1" style="ellipse;whiteSpace=wrap;html=1;fillColor=#41A56C;strokeColor=#41A56C;fontColor=#FFFFFF;fontSize=18;fontStyle=1;align=center;verticalAlign=middle;" value="H" vertex="1">
+          <mxGeometry height="72.85" width="76.62" x="1240" y="132.82" as="geometry" />
+        </mxCell>
+        <mxCell id="hospitalTitle" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=28;fontStyle=1;fontColor=#152641;align=left;" value="Hospital Portal" vertex="1">
+          <mxGeometry height="51.60000000000001" width="370.2013422818792" x="1339.0939597315437" y="144.96470588235297" as="geometry" />
+        </mxCell>
+        <mxCell id="hospitalBody" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=20;fontColor=#5E708C;align=left;verticalAlign=top;" value="Available patients, search, and trial view&lt;br&gt;Connections, inbox chat, and profile tools" vertex="1">
+          <mxGeometry height="69.81176470588235" width="462.751677852349" x="1339.0939597315437" y="205.67058823529413" as="geometry" />
+        </mxCell>
+        <mxCell id="flaskCard" parent="1" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#F3EEFF;strokeColor=#7A5CD3;strokeWidth=3;arcSize=18;shadow=1;" value="" vertex="1">
+          <mxGeometry height="265.5882352941176" width="819.7315436241612" x="644.9664429530202" y="375.6470588235295" as="geometry" />
+        </mxCell>
+        <mxCell id="flaskBadge" parent="1" style="ellipse;whiteSpace=wrap;html=1;fillColor=#7A5CD3;strokeColor=#7A5CD3;fontColor=#FFFFFF;fontSize=18;fontStyle=1;align=center;verticalAlign=middle;" value="W" vertex="1">
+          <mxGeometry height="78.92" width="81.89" x="700" y="419.38" as="geometry" />
+        </mxCell>
+        <mxCell id="flaskTitle" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=30;fontStyle=1;fontColor=#152641;align=left;" value="Flask Web Application" vertex="1">
+          <mxGeometry height="57.67058823529412" width="502.41610738255036" x="814.203422818792" y="430.00352941176465" as="geometry" />
+        </mxCell>
+        <mxCell id="flaskBody" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=20;fontColor=#5E708C;align=left;verticalAlign=top;" value="Renders patient and hospital pages&lt;br&gt;Handles login, APIs, sessions, and requests&lt;br&gt;Coordinates matching, inbox, connections, and uploads" vertex="1">
+          <mxGeometry height="80.8" width="402.99" x="797.01" y="509.2" as="geometry" />
+        </mxCell>
+        <mxCell id="dbCard" parent="1" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#EFF9F3;strokeColor=#39A66E;strokeWidth=3;arcSize=18;shadow=1;" value="" vertex="1">
+          <mxGeometry height="273.1764705882353" width="502.41610738255036" x="155.7718120805369" y="770.2352941176471" as="geometry" />
+        </mxCell>
+        <mxCell id="dbBadge" parent="1" style="ellipse;whiteSpace=wrap;html=1;fillColor=#39A66E;strokeColor=#39A66E;fontColor=#FFFFFF;fontSize=18;fontStyle=1;align=center;verticalAlign=middle;" value="D" vertex="1">
+          <mxGeometry height="72.85" width="78.73" x="195.44" y="815.77" as="geometry" />
+        </mxCell>
+        <mxCell id="dbTitle" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=28;fontStyle=1;fontColor=#152641;align=left;" value="App Database + Files" vertex="1">
+          <mxGeometry height="51.60000000000001" width="383.4228187919463" x="289.9980536912752" y="826.3858823529412" as="geometry" />
+        </mxCell>
+        <mxCell id="dbBody" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=20;fontColor=#5E708C;align=left;verticalAlign=top;" value="SQLite for accounts, connections, and messages&lt;br&gt;Uploads folder for patient documents&lt;br&gt;Keeps local portal state for the demo" vertex="1">
+          <mxGeometry height="109.27058823529411" width="383.4228187919463" x="202.00993288590604" y="910.0011764705881" as="geometry" />
+        </mxCell>
+        <mxCell id="mlCard" parent="1" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#FFF5E5;strokeColor=#DD8E1D;strokeWidth=3;arcSize=18;shadow=1;" value="" vertex="1">
+          <mxGeometry height="273.1764705882353" width="502.41610738255036" x="803.6241610738255" y="770.2352941176471" as="geometry" />
+        </mxCell>
+        <mxCell id="mlBadge" parent="1" style="ellipse;whiteSpace=wrap;html=1;fillColor=#DD8E1D;strokeColor=#DD8E1D;fontColor=#FFFFFF;fontSize=18;fontStyle=1;align=center;verticalAlign=middle;" value="M" vertex="1">
+          <mxGeometry height="72.85" width="77.46" x="852.54" y="815.76" as="geometry" />
+        </mxCell>
+        <mxCell id="mlTitle" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=28;fontStyle=1;fontColor=#152641;align=left;" value="Trial Matching Engine" vertex="1">
+          <mxGeometry height="51.6" width="310" x="960" y="826.39" as="geometry" />
+        </mxCell>
+        <mxCell id="mlBody" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=20;fontColor=#5E708C;align=left;verticalAlign=top;" value="Pipeline loads patient and trial datasets&lt;br&gt;Feature engineering + Random Forest model&lt;br&gt;Returns ranked trial matches and hospital trial lists" vertex="1">
+          <mxGeometry height="109.27058823529411" width="403.255033557047" x="852.5422818791948" y="910.0011764705881" as="geometry" />
+        </mxCell>
+        <mxCell id="srcCard" parent="1" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#EDF9F9;strokeColor=#2D9FA1;strokeWidth=3;arcSize=18;shadow=1;" value="" vertex="1">
+          <mxGeometry height="273.1764705882353" width="502.41610738255036" x="1451.476510067114" y="770.2352941176471" as="geometry" />
+        </mxCell>
+        <mxCell id="srcBadge" parent="1" style="ellipse;whiteSpace=wrap;html=1;fillColor=#2D9FA1;strokeColor=#2D9FA1;fontColor=#FFFFFF;fontSize=18;fontStyle=1;align=center;verticalAlign=middle;" value="S" vertex="1">
+          <mxGeometry height="72.85" width="78.4" x="1491.61" y="815.76" as="geometry" />
+        </mxCell>
+        <mxCell id="srcTitle" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=28;fontStyle=1;fontColor=#152641;align=left;" value="Source Data" vertex="1">
+          <mxGeometry height="51.60000000000001" width="290.8724832214765" x="1600.0027516778523" y="826.3858823529412" as="geometry" />
+        </mxCell>
+        <mxCell id="srcBody" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=20;fontColor=#5E708C;align=left;verticalAlign=top;" value="Synthea patient CSVs&lt;br&gt;ClinicalTrials.gov / AACT trial CSVs&lt;br&gt;MIMIC demo data for validation" vertex="1">
+          <mxGeometry height="109.27058823529411" width="383.4228187919463" x="1507.4546308724832" y="897.7211764705881" as="geometry" />
+        </mxCell>
+        <mxCell id="lblUser" parent="1" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#FFFFFF;strokeColor=none;fontSize=18;fontStyle=1;fontColor=#5C6E89;align=center;verticalAlign=middle;" value="user requests" vertex="1">
+          <mxGeometry height="48.56470588235294" width="156.01342281879195" x="976.8232214765102" y="287.6205882352941" as="geometry" />
+        </mxCell>
+        <mxCell id="lblRead" parent="1" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#FFFFFF;strokeColor=none;fontSize=18;fontStyle=1;fontColor=#5C6E89;align=center;verticalAlign=middle;" value="read / write" vertex="1">
+          <mxGeometry height="48.56470588235294" width="145.43624161073825" x="439.9987919463087" y="660.0011764705882" as="geometry" />
+        </mxCell>
+        <mxCell id="lblMatch" parent="1" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#FFFFFF;strokeColor=none;fontSize=18;fontStyle=1;fontColor=#5C6E89;align=center;verticalAlign=middle;" value="match" vertex="1">
+          <mxGeometry height="48.56470588235294" width="105.7718120805369" x="930.0036912751677" y="690.0011764705882" as="geometry" />
+        </mxCell>
+        <mxCell id="lblRank" parent="1" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#FFFFFF;strokeColor=none;fontSize=18;fontStyle=1;fontColor=#5C6E89;align=center;verticalAlign=middle;" value="ranked results" vertex="1">
+          <mxGeometry height="48.56470588235294" width="166.59060402684565" x="1150.029932885906" y="689.9982352941176" as="geometry" />
+        </mxCell>
+        <mxCell id="lblFeed" parent="1" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#FFFFFF;strokeColor=none;fontSize=18;fontStyle=1;fontColor=#5C6E89;align=center;verticalAlign=middle;" value="data feed" vertex="1">
+          <mxGeometry height="48.56470588235294" width="137.50335570469798" x="1633.9364429530203" y="697.3929411764708" as="geometry" />
+        </mxCell>
+        <mxCell id="e1" edge="1" parent="1" source="patientCard" style="edgeStyle=orthogonalEdgeStyle;rounded=1;orthogonalLoop=1;jettySize=auto;html=1;endArrow=block;endFill=1;strokeColor=#42536F;strokeWidth=3;" target="flaskCard" value="">
+          <mxGeometry relative="1" as="geometry">
+            <Array as="points">
+              <mxPoint x="730" y="320" />
+              <mxPoint x="730" y="320" />
+            </Array>
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="e2" edge="1" parent="1" source="hospitalCard" style="edgeStyle=orthogonalEdgeStyle;rounded=1;orthogonalLoop=1;jettySize=auto;html=1;endArrow=block;endFill=1;strokeColor=#42536F;strokeWidth=3;" target="flaskCard" value="">
+          <mxGeometry relative="1" as="geometry">
+            <Array as="points">
+              <mxPoint x="1380" y="330" />
+              <mxPoint x="1380" y="330" />
+            </Array>
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="e4" edge="1" parent="1" source="flaskCard" style="edgeStyle=orthogonalEdgeStyle;rounded=1;orthogonalLoop=1;jettySize=auto;html=1;endArrow=block;endFill=1;strokeColor=#42536F;strokeWidth=3;" target="mlCard" value="">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="e5" edge="1" parent="1" source="srcCard" style="edgeStyle=orthogonalEdgeStyle;rounded=1;orthogonalLoop=1;jettySize=auto;html=1;endArrow=block;endFill=1;strokeColor=#42536F;strokeWidth=3;" target="mlCard" value="">
+          <mxGeometry relative="1" as="geometry" />
+        </mxCell>
+        <mxCell id="e6" edge="1" parent="1" source="mlCard" style="edgeStyle=orthogonalEdgeStyle;rounded=1;orthogonalLoop=1;jettySize=auto;html=1;endArrow=block;endFill=1;strokeColor=#42536F;strokeWidth=3;" target="flaskCard" value="">
+          <mxGeometry relative="1" as="geometry">
+            <Array as="points">
+              <mxPoint x="1130" y="700" />
+              <mxPoint x="1130" y="700" />
+            </Array>
+          </mxGeometry>
+        </mxCell>
+        <mxCell id="footer" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=18;fontStyle=1;fontColor=#667892;align=center;" value="FLOW: portals -&amp;gt; Flask app -&amp;gt; matching engine -&amp;gt; recommendations, with app state stored in SQLite" vertex="1">
+          <mxGeometry height="36.423529411764704" width="1322.1476510067116" x="393.758389261745" y="1113.2235294117647" as="geometry" />
+        </mxCell>
+        <mxCell id="sec2" parent="1" style="text;html=1;strokeColor=none;fillColor=none;fontSize=18;fontStyle=1;fontColor=#61728E;letterSpacing=1;" value="APPLICATION LAYER" vertex="1">
+          <mxGeometry height="36.423529411764704" width="251.20805369127518" x="499.9986577181208" y="319.99823529411765" as="geometry" />
+        </mxCell>
+        <mxCell id="7qRbrwoMCzgDvgHPFRhy-5" edge="1" parent="1" style="edgeStyle=orthogonalEdgeStyle;rounded=1;orthogonalLoop=1;jettySize=auto;html=1;endArrow=block;endFill=1;strokeColor=#42536F;strokeWidth=3;entryX=0.5;entryY=0;entryDx=0;entryDy=0;" target="dbCard" value="">
+          <mxGeometry relative="1" as="geometry">
+            <Array as="points">
+              <mxPoint x="407" y="550" />
+            </Array>
+            <mxPoint x="640" y="550" as="sourcePoint" />
+            <mxPoint x="400" y="709" as="targetPoint" />
+          </mxGeometry>
+        </mxCell>
+      </root>
+    </mxGraphModel>
+  </diagram>
+</mxfile>

Architecture Diagrams/secondlife low level architecture.drawio

@@ -0,0 +1,306 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<mxGraphModel dx="1422" dy="762" grid="1" gridSize="10" guides="1" tooltips="1" connect="1" arrows="1" fold="1" page="1" pageScale="1" pageWidth="1700" pageHeight="1100" math="0" shadow="0">
+  <root>
+    <mxCell id="0" />
+    <mxCell id="1" parent="0" />
+
+    <!-- ============================================================
+         TITLE
+         ============================================================ -->
+    <mxCell id="TITLE" value="Second Life — System Architecture  |  DSCI 5260 · Group 7  |  AI-Powered Clinical Trial Matching" style="text;html=1;strokeColor=none;fillColor=none;fontSize=16;fontStyle=1;fontColor=#003366;align=center;" vertex="1" parent="1">
+      <mxGeometry x="10" y="1078" width="1680" height="30" as="geometry" />
+    </mxCell>
+
+    <!-- ============================================================
+         LAYER 1 — BROWSER
+         ============================================================ -->
+    <mxCell id="L1" value="🌐  Browser  (Client)" style="swimlane;startSize=32;fillColor=#dae8fc;strokeColor=#6c8ebf;fontSize=14;fontStyle=1;fontColor=#003366;swimlaneLine=1;" vertex="1" parent="1">
+      <mxGeometry x="10" y="10" width="1680" height="235" as="geometry" />
+    </mxCell>
+
+    <!-- Patient Portal -->
+    <mxCell id="PPG" value="Patient Portal — patient.html" style="swimlane;startSize=24;fillColor=#cce5ff;strokeColor=#0066cc;fontSize=11;fontStyle=1;" vertex="1" parent="L1">
+      <mxGeometry x="20" y="37" width="574" height="183" as="geometry" />
+    </mxCell>
+    <mxCell id="PP1" value="My Profile&#xa;conditions · meds&#xa;documents upload" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#e8f4fd;strokeColor=#0066cc;fontSize=9;" vertex="1" parent="PPG">
+      <mxGeometry x="10" y="30" width="120" height="65" as="geometry" />
+    </mxCell>
+    <mxCell id="PP2" value="Suggested Trials&#xa;AI-ranked top-20&#xa;combined_score" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#e8f4fd;strokeColor=#0066cc;fontSize=9;" vertex="1" parent="PPG">
+      <mxGeometry x="140" y="30" width="120" height="65" as="geometry" />
+    </mxCell>
+    <mxCell id="PP3" value="My Connections&#xa;status · chat modal&#xa;view history" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#e8f4fd;strokeColor=#0066cc;fontSize=9;" vertex="1" parent="PPG">
+      <mxGeometry x="270" y="30" width="120" height="65" as="geometry" />
+    </mxCell>
+    <mxCell id="PP4" value="Inbox 🔴&#xa;unread count badge&#xa;all message threads" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#e8f4fd;strokeColor=#0066cc;fontSize=9;" vertex="1" parent="PPG">
+      <mxGeometry x="400" y="30" width="120" height="65" as="geometry" />
+    </mxCell>
+    <mxCell id="PP_NOTE" value="Bootstrap 5.3 dark SPA  ·  escH() / escA() XSS guards  ·  DOM API only (no innerHTML for user data)  ·  startStatusPoll() 5s interval  ·  cookie session auth" style="text;html=1;strokeColor=none;fillColor=none;fontSize=8;fontColor=#555;align=left;" vertex="1" parent="PPG">
+      <mxGeometry x="10" y="106" width="554" height="16" as="geometry" />
+    </mxCell>
+
+    <!-- Landing Page -->
+    <mxCell id="LP" value="landing.html&#xa;&#xa;Login · Register&#xa;Patient &amp; Hospital" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#e6f3ff;strokeColor=#3385ff;fontSize=10;fontStyle=1;" vertex="1" parent="L1">
+      <mxGeometry x="651" y="70" width="158" height="95" as="geometry" />
+    </mxCell>
+
+    <!-- Hospital Portal -->
+    <mxCell id="HPG" value="Hospital Portal — hospital.html" style="swimlane;startSize=24;fillColor=#d5e8d4;strokeColor=#5b9e5b;fontSize=11;fontStyle=1;" vertex="1" parent="L1">
+      <mxGeometry x="866" y="37" width="804" height="183" as="geometry" />
+    </mxCell>
+    <mxCell id="HP1" value="Available&#xa;Patients&#xa;opt-in only" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#e8f5e8;strokeColor=#5b9e5b;fontSize=9;" vertex="1" parent="HPG">
+      <mxGeometry x="10" y="30" width="113" height="65" as="geometry" />
+    </mxCell>
+    <mxCell id="HP2" value="Search by&#xa;Condition&#xa;incl. connected" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#e8f5e8;strokeColor=#5b9e5b;fontSize=9;" vertex="1" parent="HPG">
+      <mxGeometry x="133" y="30" width="113" height="65" as="geometry" />
+    </mxCell>
+    <mxCell id="HP3" value="My Trials&#xa;3-tier active&#xa;match" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#e8f5e8;strokeColor=#5b9e5b;fontSize=9;" vertex="1" parent="HPG">
+      <mxGeometry x="256" y="30" width="113" height="65" as="geometry" />
+    </mxCell>
+    <mxCell id="HP4" value="Inbox 🔴&#xa;unread badge&#xa;message threads" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#e8f5e8;strokeColor=#5b9e5b;fontSize=9;" vertex="1" parent="HPG">
+      <mxGeometry x="379" y="30" width="113" height="65" as="geometry" />
+    </mxCell>
+    <mxCell id="HP5" value="My&#xa;Connections&#xa;accept / reject" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#e8f5e8;strokeColor=#5b9e5b;fontSize=9;" vertex="1" parent="HPG">
+      <mxGeometry x="502" y="30" width="113" height="65" as="geometry" />
+    </mxCell>
+    <mxCell id="HP6" value="My Profile&#xa;name · location&#xa;conditions" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#e8f5e8;strokeColor=#5b9e5b;fontSize=9;" vertex="1" parent="HPG">
+      <mxGeometry x="625" y="30" width="113" height="65" as="geometry" />
+    </mxCell>
+    <mxCell id="HP_NOTE" value="Bootstrap 5.3 dark SPA  ·  escH() / escA() XSS guards  ·  navbar name updates live on profile save  ·  Search includes already-connected toggle" style="text;html=1;strokeColor=none;fillColor=none;fontSize=8;fontColor=#555;align=left;" vertex="1" parent="HPG">
+      <mxGeometry x="10" y="106" width="784" height="16" as="geometry" />
+    </mxCell>
+
+    <!-- ============================================================
+         LAYER 2 — FLASK SERVER
+         ============================================================ -->
+    <mxCell id="L2" value="⚙️  Flask Server — app.py  (port 5000)  ·  TEMPLATES_AUTO_RELOAD=True  ·  MAX_CONTENT_LENGTH=10 MB  ·  server-side cookie sessions" style="swimlane;startSize=32;fillColor=#e1d5e7;strokeColor=#9673a6;fontSize=13;fontStyle=1;fontColor=#3d0066;" vertex="1" parent="1">
+      <mxGeometry x="10" y="255" width="1680" height="280" as="geometry" />
+    </mxCell>
+
+    <mxCell id="AR" value="🔐 Auth Routes&#xa;(no session required)&#xa;&#xa;POST /auth/patient/register&#xa;POST /auth/patient/login&#xa;POST /auth/hospital/register&#xa;POST /auth/hospital/login&#xa;POST /auth/logout" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#f3e5f5;strokeColor=#9673a6;fontSize=9;align=left;spacingLeft=8;verticalAlign=top;spacingTop=8;" vertex="1" parent="L2">
+      <mxGeometry x="20" y="40" width="215" height="220" as="geometry" />
+    </mxCell>
+
+    <mxCell id="PAR" value="👤 Patient API&#xa;(patient session)&#xa;&#xa;GET | POST  /api/patient/profile&#xa;GET  /api/patient/matches&#xa;GET | POST | DELETE&#xa;     /api/patient/interest&#xa;GET  /api/patient/connections&#xa;POST  /api/patient/connect&#xa;GET  /api/patient/hospitals-for-trial" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#cce5ff;strokeColor=#0066cc;fontSize=9;align=left;spacingLeft=8;verticalAlign=top;spacingTop=8;" vertex="1" parent="L2">
+      <mxGeometry x="248" y="40" width="250" height="220" as="geometry" />
+    </mxCell>
+
+    <mxCell id="HAR" value="🏥 Hospital API&#xa;(hospital session)&#xa;&#xa;GET | POST  /api/hospital/profile&#xa;GET  /api/hospital/patients&#xa;GET  /api/hospital/trials&#xa;POST  /api/hospital/connect&#xa;GET  /api/hospital/connections&#xa;PUT  /api/hospital/connections/&#xa;       &lt;cid&gt;/status" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#d5e8d4;strokeColor=#5b9e5b;fontSize=9;align=left;spacingLeft=8;verticalAlign=top;spacingTop=8;" vertex="1" parent="L2">
+      <mxGeometry x="511" y="40" width="265" height="220" as="geometry" />
+    </mxCell>
+
+    <mxCell id="MR" value="💬 Messaging + Inbox&#xa;&#xa;GET | POST&#xa;  /api/patient/connections/&lt;cid&gt;/messages&#xa;GET  /api/patient/inbox&#xa;&#xa;GET | POST&#xa;  /api/hospital/connections/&lt;cid&gt;/messages&#xa;GET  /api/hospital/inbox&#xa;&#xa;→ mark_messages_read() on GET&#xa;→ get_*_inbox_threads() + unread_count" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#fff2cc;strokeColor=#d6b656;fontSize=9;align=left;spacingLeft=8;verticalAlign=top;spacingTop=8;" vertex="1" parent="L2">
+      <mxGeometry x="789" y="40" width="280" height="220" as="geometry" />
+    </mxCell>
+
+    <mxCell id="DR" value="📄 Documents&#xa;&#xa;GET  /api/patient/documents&#xa;POST  /api/patient/documents&#xa;  → multipart/form-data&#xa;  → secure_filename()&#xa;  → ALLOWED: .pdf .docx .doc&#xa;             .txt .png .jpg .jpeg&#xa;  → max 10 MB&#xa;DELETE  /api/patient/documents/&lt;id&gt;&#xa;GET  /api/patient/documents/&lt;id&gt;/download" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#ffe6cc;strokeColor=#d79b00;fontSize=9;align=left;spacingLeft=8;verticalAlign=top;spacingTop=8;" vertex="1" parent="L2">
+      <mxGeometry x="1082" y="40" width="255" height="220" as="geometry" />
+    </mxCell>
+
+    <mxCell id="SR" value="🔧 Shared + Boot Thread&#xa;&#xa;GET /api/status&#xa;  → ready: true|false + stats&#xa;GET /api/conditions/autocomplete&#xa;&#xa;─────────────────────&#xa;🚀 Boot Thread (background)&#xa;① db.init_db()&#xa;  → create 5 tables&#xa;  → seed 25 patients + 3 hospitals&#xa;② pipeline.load()&#xa;  → read 11 CSV files into memory&#xa;③ pipeline.train()&#xa;  → build RF or load model_cache.pkl" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#f5f5f5;strokeColor=#666666;fontSize=9;align=left;spacingLeft=8;verticalAlign=top;spacingTop=8;" vertex="1" parent="L2">
+      <mxGeometry x="1350" y="40" width="310" height="220" as="geometry" />
+    </mxCell>
+
+    <!-- ============================================================
+         LAYER 3 — BUSINESS LOGIC
+         ============================================================ -->
+    <mxCell id="L3" value="🧠  Business Logic" style="swimlane;startSize=32;fillColor=#fff8e1;strokeColor=#d6a000;fontSize=14;fontStyle=1;fontColor=#5a3e00;" vertex="1" parent="1">
+      <mxGeometry x="10" y="545" width="1680" height="300" as="geometry" />
+    </mxCell>
+
+    <!-- database.py -->
+    <mxCell id="DBG" value="database.py — SQLite Layer" style="swimlane;startSize=24;fillColor=#d5e8d4;strokeColor=#5b9e5b;fontSize=11;fontStyle=1;" vertex="1" parent="L3">
+      <mxGeometry x="20" y="38" width="705" height="252" as="geometry" />
+    </mxCell>
+    <mxCell id="T1" value="patient_accounts&#xa;25 rows&#xa;5 demo + 20 Synthea" style="shape=cylinder3;whiteSpace=wrap;html=1;boundedLbl=1;backgroundOutline=1;size=12;fillColor=#d5e8d4;strokeColor=#5b9e5b;fontSize=9;" vertex="1" parent="DBG">
+      <mxGeometry x="10" y="32" width="122" height="68" as="geometry" />
+    </mxCell>
+    <mxCell id="T2" value="hospital_accounts&#xa;3 rows&#xa;mgh, cleveland, jhopkins" style="shape=cylinder3;whiteSpace=wrap;html=1;boundedLbl=1;backgroundOutline=1;size=12;fillColor=#d5e8d4;strokeColor=#5b9e5b;fontSize=9;" vertex="1" parent="DBG">
+      <mxGeometry x="144" y="32" width="122" height="68" as="geometry" />
+    </mxCell>
+    <mxCell id="T3" value="patient_trial_interests&#xa;UNIQUE(patient_id,&#xa;trial_id)" style="shape=cylinder3;whiteSpace=wrap;html=1;boundedLbl=1;backgroundOutline=1;size=12;fillColor=#d5e8d4;strokeColor=#5b9e5b;fontSize=9;" vertex="1" parent="DBG">
+      <mxGeometry x="278" y="32" width="122" height="68" as="geometry" />
+    </mxCell>
+    <mxCell id="T4" value="connections&#xa;UNIQUE(patient, hospital,&#xa;COALESCE(trial,''))" style="shape=cylinder3;whiteSpace=wrap;html=1;boundedLbl=1;backgroundOutline=1;size=12;fillColor=#d5e8d4;strokeColor=#5b9e5b;fontSize=9;" vertex="1" parent="DBG">
+      <mxGeometry x="412" y="32" width="132" height="68" as="geometry" />
+    </mxCell>
+    <mxCell id="T5" value="connection_messages&#xa;sender_role: patient|hospital&#xa;is_read: 0 | 1" style="shape=cylinder3;whiteSpace=wrap;html=1;boundedLbl=1;backgroundOutline=1;size=12;fillColor=#d5e8d4;strokeColor=#5b9e5b;fontSize=9;" vertex="1" parent="DBG">
+      <mxGeometry x="556" y="32" width="132" height="68" as="geometry" />
+    </mxCell>
+    <mxCell id="DBF1" value="Key functions:  authenticate_patient / authenticate_hospital  ·  update_patient_profile / update_hospital_profile  ·  get_open_patients_for_hospital(condition, include_connected)  ·  create_connection / connection_exists  ·  update_connection_status  ·  create_connection_message  ·  mark_messages_read  ·  get_hospital_inbox_threads / get_patient_inbox_threads  ·  _seed_dataset_patients(20)" style="text;html=1;strokeColor=none;fillColor=none;fontSize=8;fontColor=#333;align=left;" vertex="1" parent="DBG">
+      <mxGeometry x="10" y="112" width="685" height="30" as="geometry" />
+    </mxCell>
+    <mxCell id="DBF2" value="JSON columns (conditions, medications, documents, research_conditions) stored as TEXT  ·  parsed by _row_to_dict()  ·  passwords: SHA-256 via _hash()  ·  idx_conn_unique handles NULL trial_id  ·  idx_msgs_conn on (connection_id, created_at)" style="text;html=1;strokeColor=none;fillColor=none;fontSize=8;fontColor=#555;align=left;" vertex="1" parent="DBG">
+      <mxGeometry x="10" y="148" width="685" height="30" as="geometry" />
+    </mxCell>
+
+    <!-- pipeline.py -->
+    <mxCell id="PLG" value="pipeline.py — SecondLifePipeline (ML Engine)" style="swimlane;startSize=24;fillColor=#ffe6cc;strokeColor=#d79b00;fontSize=11;fontStyle=1;" vertex="1" parent="L3">
+      <mxGeometry x="745" y="38" width="925" height="252" as="geometry" />
+    </mxCell>
+    <mxCell id="PL1" value="load()&#xa;Reads 11 CSV files → memory&#xa;trial_profiles · active_trial_ids&#xa;trial_us_states · facility_tokens&#xa;drug_keywords · summaries&#xa;patient_conditions · meds · labs&#xa;~2–3 min on first boot" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#fff3e0;strokeColor=#d79b00;fontSize=9;" vertex="1" parent="PLG">
+      <mxGeometry x="10" y="32" width="165" height="115" as="geometry" />
+    </mxCell>
+    <mxCell id="PL2" value="train()&#xa;3000 patients × 30 trials&#xa;_compute_features() 17 feats&#xa;pseudo-label: 6-feat weighted score&#xa;GroupShuffleSplit 80/20 (patient-level)&#xa;RandomForest n=200, depth=12&#xa;CalibratedClassifierCV isotonic cv=3&#xa;→ save model_cache.pkl" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#fff3e0;strokeColor=#d79b00;fontSize=9;" vertex="1" parent="PLG">
+      <mxGeometry x="188" y="32" width="180" height="115" as="geometry" />
+    </mxCell>
+    <mxCell id="PL3" value="match_patient()&#xa;Input: conditions, age,&#xa;        gender, patient_id&#xa;Scores all active trials&#xa;combined = 0.6 × RF_prob&#xa;          + 0.4 × match_score&#xa;Returns top-20 ranked trials" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#fff3e0;strokeColor=#d79b00;fontSize=9;" vertex="1" parent="PLG">
+      <mxGeometry x="382" y="32" width="170" height="115" as="geometry" />
+    </mxCell>
+    <mxCell id="PL4" value="trials_for_hospital()&#xa;Input: name, location,&#xa;        research_conditions&#xa;Tier1: facility Jaccard ≥ 0.25&#xa;Tier2: same US state&#xa;Tier3: condition overlap&#xa;Active-only filter every tier" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#fff3e0;strokeColor=#d79b00;fontSize=9;" vertex="1" parent="PLG">
+      <mxGeometry x="566" y="32" width="165" height="115" as="geometry" />
+    </mxCell>
+    <mxCell id="PL5" value="hospitals_for_trial()&#xa;Input: trial_id&#xa;Tier1: Jaccard ≥ 0.25 → 🟢&#xa;Tier2: same state → ⬜&#xa;Tier3: cond overlap → ⬜&#xa;Tier4: fallback → ⬜&#xa;Modal: green vs grey sections" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#fff3e0;strokeColor=#d79b00;fontSize=9;" vertex="1" parent="PLG">
+      <mxGeometry x="745" y="32" width="165" height="115" as="geometry" />
+    </mxCell>
+    <mxCell id="PLM" value="17 Features: condition_overlap · jaccard_similarity · overlap_ratio_trial · overlap_ratio_patient · condition_rarity_score · trial_specificity · condition_burden · active_ratio · resolved_ratio · age_distance · age_centered · age_compatibility · gender_compatibility · geo_feasibility · med_compatibility · lab_availability · data_completeness" style="text;html=1;strokeColor=none;fillColor=none;fontSize=8;fontColor=#333;align=left;" vertex="1" parent="PLG">
+      <mxGeometry x="10" y="158" width="905" height="20" as="geometry" />
+    </mxCell>
+    <mxCell id="PLM2" value="AUC-ROC: 0.60  ·  Accuracy: 85.1%  ·  F1: 0.917  ·  Precision: 0.852  ·  Recall: 0.993  ·  90,994 training pairs  ·  82.1% pseudo-label positive rate  ·  GroupShuffleSplit prevents patient-level data leakage" style="text;html=1;strokeColor=none;fillColor=none;fontSize=8;fontColor=#555;align=left;" vertex="1" parent="PLG">
+      <mxGeometry x="10" y="182" width="905" height="20" as="geometry" />
+    </mxCell>
+
+    <!-- ============================================================
+         LAYER 4 — STORAGE
+         ============================================================ -->
+    <mxCell id="L4" value="💾  Storage" style="swimlane;startSize=32;fillColor=#f5f5f5;strokeColor=#666666;fontSize=14;fontStyle=1;fontColor=#333;" vertex="1" parent="1">
+      <mxGeometry x="10" y="855" width="1680" height="215" as="geometry" />
+    </mxCell>
+
+    <mxCell id="S1" value="secondlife.db&#xa;SQLite file&#xa;5 tables · auto-created&#xa;on init_db()" style="shape=cylinder3;whiteSpace=wrap;html=1;boundedLbl=1;backgroundOutline=1;size=14;fillColor=#d5e8d4;strokeColor=#5b9e5b;fontStyle=1;fontSize=10;" vertex="1" parent="L4">
+      <mxGeometry x="75" y="45" width="155" height="120" as="geometry" />
+    </mxCell>
+
+    <mxCell id="S2" value="model_cache.pkl&#xa;RandomForest +&#xa;CalibratedClassifierCV&#xa;delete to force retrain" style="shape=cylinder3;whiteSpace=wrap;html=1;boundedLbl=1;backgroundOutline=1;size=14;fillColor=#ffe6cc;strokeColor=#d79b00;fontStyle=1;fontSize=10;" vertex="1" parent="L4">
+      <mxGeometry x="325" y="45" width="155" height="120" as="geometry" />
+    </mxCell>
+
+    <mxCell id="S3" value="uploads/patient_docs/&#xa;&lt;patient_id&gt;/&#xa;&lt;doc_id&gt;_filename&#xa;10 MB max · auto-created" style="shape=cylinder3;whiteSpace=wrap;html=1;boundedLbl=1;backgroundOutline=1;size=14;fillColor=#cce5ff;strokeColor=#0066cc;fontStyle=1;fontSize=10;" vertex="1" parent="L4">
+      <mxGeometry x="575" y="45" width="155" height="120" as="geometry" />
+    </mxCell>
+
+    <mxCell id="S4" value="Synthea Patient CSVs&#xa;&#xa;patients_details.csv&#xa;  265,893 patients · demographics&#xa;final_patients_conditions.csv&#xa;  967k rows · 106 conditions&#xa;patients_medications.csv&#xa;  213,182 patients&#xa;patients_observations.csv&#xa;  23,231 patients with lab data" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#e8f5e8;strokeColor=#5b9e5b;fontSize=9;align=left;spacingLeft=8;verticalAlign=top;spacingTop=8;" vertex="1" parent="L4">
+      <mxGeometry x="855" y="38" width="275" height="158" as="geometry" />
+    </mxCell>
+
+    <mxCell id="S5" value="ClinicalTrials.gov / AACT CSVs&#xa;&#xa;trail_studies.csv — 65k recruiting trials&#xa;trail_conditions.csv — 34k with conditions&#xa;trail_eligibilities.csv — age · gender&#xa;trail_facilities.csv — 189k with US geo&#xa;trail_interventions.csv — 196k with drugs&#xa;trail_brief_summaries.csv — summaries&#xa;trail_keywords.csv — keyword index" style="rounded=1;whiteSpace=wrap;html=1;fillColor=#fff3e0;strokeColor=#d79b00;fontSize=9;align=left;spacingLeft=8;verticalAlign=top;spacingTop=8;" vertex="1" parent="L4">
+      <mxGeometry x="1230" y="38" width="310" height="158" as="geometry" />
+    </mxCell>
+
+    <!-- ============================================================
+         LEGEND
+         ============================================================ -->
+    <mxCell id="LEG" value="Legend" style="swimlane;startSize=22;fillColor=#ffffff;strokeColor=#999999;fontSize=10;fontStyle=1;" vertex="1" parent="1">
+      <mxGeometry x="10" y="1085" width="1680" height="0" as="geometry" />
+    </mxCell>
+
+    <!-- ============================================================
+         ARROWS — DATA FLOW
+         ============================================================ -->
+
+    <!-- Patient Portal → Patient API (REST/JSON) -->
+    <mxCell id="E1" value="REST/JSON" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#0066cc;strokeWidth=2;fontStyle=2;fontSize=8;exitX=0.5;exitY=1;exitDx=0;exitDy=0;entryX=0.5;entryY=0;entryDx=0;entryDy=0;" edge="1" source="PPG" target="PAR" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Patient Portal → Messaging/Inbox (dashed) -->
+    <mxCell id="E2" value="chat / inbox" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#d6b656;strokeWidth=1.5;dashed=1;fontStyle=2;fontSize=8;" edge="1" source="PP4" target="MR" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Patient Portal → Documents (dashed) -->
+    <mxCell id="E3" value="upload / download" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#d79b00;strokeWidth=1.5;dashed=1;fontStyle=2;fontSize=8;" edge="1" source="PP1" target="DR" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Landing → Auth -->
+    <mxCell id="E4" value="login / register" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#9673a6;strokeWidth=2;fontStyle=2;fontSize=8;" edge="1" source="LP" target="AR" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Hospital Portal → Hospital API -->
+    <mxCell id="E5" value="REST/JSON" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#5b9e5b;strokeWidth=2;fontStyle=2;fontSize=8;exitX=0.5;exitY=1;exitDx=0;exitDy=0;entryX=0.5;entryY=0;entryDx=0;entryDy=0;" edge="1" source="HPG" target="HAR" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Hospital Portal → Messaging (dashed) -->
+    <mxCell id="E6" value="chat / inbox" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#d6b656;strokeWidth=1.5;dashed=1;fontStyle=2;fontSize=8;" edge="1" source="HP4" target="MR" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Auth → DB -->
+    <mxCell id="E7" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#9673a6;strokeWidth=1.5;exitX=0.5;exitY=1;exitDx=0;exitDy=0;entryX=0.1;entryY=0;entryDx=0;entryDy=0;" edge="1" source="AR" target="DBG" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Patient API → DB -->
+    <mxCell id="E8" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#0066cc;strokeWidth=2;exitX=0.5;exitY=1;exitDx=0;exitDy=0;entryX=0.3;entryY=0;entryDx=0;entryDy=0;" edge="1" source="PAR" target="DBG" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Patient API → match_patient() -->
+    <mxCell id="E9" value="match_patient()" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#d79b00;strokeWidth=2;fontStyle=3;fontSize=8;exitX=0.5;exitY=1;exitDx=0;exitDy=0;entryX=0.5;entryY=0;entryDx=0;entryDy=0;" edge="1" source="PAR" target="PL3" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Patient API → hospitals_for_trial() -->
+    <mxCell id="E10" value="hospitals_for_trial()" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#d79b00;strokeWidth=1.5;dashed=1;fontStyle=3;fontSize=8;exitX=0.75;exitY=1;exitDx=0;exitDy=0;entryX=0.5;entryY=0;entryDx=0;entryDy=0;" edge="1" source="PAR" target="PL5" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Hospital API → DB -->
+    <mxCell id="E11" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#5b9e5b;strokeWidth=2;exitX=0.5;exitY=1;exitDx=0;exitDy=0;entryX=0.5;entryY=0;entryDx=0;entryDy=0;" edge="1" source="HAR" target="DBG" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Hospital API → trials_for_hospital() -->
+    <mxCell id="E12" value="trials_for_hospital()" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#d79b00;strokeWidth=2;fontStyle=3;fontSize=8;exitX=0.5;exitY=1;exitDx=0;exitDy=0;entryX=0.5;entryY=0;entryDx=0;entryDy=0;" edge="1" source="HAR" target="PL4" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Messaging → DB (connections + messages tables) -->
+    <mxCell id="E13" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#d6b656;strokeWidth=1.5;exitX=0.5;exitY=1;exitDx=0;exitDy=0;entryX=0.75;entryY=0;entryDx=0;entryDy=0;" edge="1" source="MR" target="DBG" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Documents → DB (updates patient documents JSON) -->
+    <mxCell id="E14" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#d79b00;strokeWidth=1.5;exitX=0.5;exitY=1;exitDx=0;exitDy=0;entryX=0.88;entryY=0;entryDx=0;entryDy=0;" edge="1" source="DR" target="DBG" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Documents → File Storage -->
+    <mxCell id="E15" value="file I/O" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#d79b00;strokeWidth=1.5;dashed=1;fontStyle=3;fontSize=8;" edge="1" source="DR" target="S3" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Boot Thread → DB (seed data) -->
+    <mxCell id="E16" value="① seed" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#666;strokeWidth=1.5;dashed=1;fontStyle=3;fontSize=8;exitX=0.5;exitY=1;exitDx=0;exitDy=0;entryX=1;entryY=0;entryDx=0;entryDy=0;" edge="1" source="SR" target="DBG" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Boot Thread → Pipeline (load + train) -->
+    <mxCell id="E17" value="② load  ③ train" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#666;strokeWidth=1.5;dashed=1;fontStyle=3;fontSize=8;exitX=0.5;exitY=1;exitDx=0;exitDy=0;entryX=1;entryY=0;entryDx=0;entryDy=0;" edge="1" source="SR" target="PLG" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- DB → secondlife.db file -->
+    <mxCell id="E18" value="read / write" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#5b9e5b;strokeWidth=2;fontStyle=2;fontSize=8;" edge="1" source="DBG" target="S1" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- Pipeline → model_cache.pkl -->
+    <mxCell id="E19" value="load / save" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#d79b00;strokeWidth=2;fontStyle=2;fontSize=8;" edge="1" source="PLG" target="S2" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- pipeline.load() → Synthea CSVs -->
+    <mxCell id="E20" value="read on boot" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#5b9e5b;strokeWidth=1.5;dashed=1;fontStyle=3;fontSize=8;exitX=0.2;exitY=1;exitDx=0;exitDy=0;entryX=0.5;entryY=0;entryDx=0;entryDy=0;" edge="1" source="PL1" target="S4" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+    <!-- pipeline.load() → Trial CSVs -->
+    <mxCell id="E21" value="read on boot" style="edgeStyle=orthogonalEdgeStyle;html=1;rounded=1;strokeColor=#d79b00;strokeWidth=1.5;dashed=1;fontStyle=3;fontSize=8;exitX=0.8;exitY=1;exitDx=0;exitDy=0;entryX=0.5;entryY=0;entryDx=0;entryDy=0;" edge="1" source="PL1" target="S5" parent="1">
+      <mxGeometry relative="1" as="geometry" />
+    </mxCell>
+
+  </root>
+</mxGraphModel>

Architecture Diagrams/secondlife workflow architecture diagrams.html

@@ -0,0 +1,828 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="UTF-8">
+<meta name="viewport" content="width=device-width, initial-scale=1.0">
+<title>Second Life — Architecture Diagrams</title>
+<script src="https://cdn.jsdelivr.net/npm/mermaid@10/dist/mermaid.min.js"></script>
+<style>
+  * { box-sizing: border-box; margin: 0; padding: 0; }
+
+  body {
+    font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", system-ui, sans-serif;
+    background: #0d1117;
+    color: #c9d1d9;
+    padding: 2rem;
+    max-width: 1600px;
+    margin: 0 auto;
+    line-height: 1.6;
+  }
+
+  header {
+    border-bottom: 1px solid #30363d;
+    padding-bottom: 1.5rem;
+    margin-bottom: 2rem;
+  }
+
+  header h1 {
+    font-size: 2rem;
+    color: #58a6ff;
+    margin-bottom: 0.25rem;
+  }
+
+  header p { color: #8b949e; font-size: 0.95rem; }
+
+  .toc {
+    background: #161b22;
+    border: 1px solid #30363d;
+    border-radius: 8px;
+    padding: 1.25rem 1.5rem;
+    margin-bottom: 3rem;
+  }
+
+  .toc h3 { color: #79c0ff; font-size: 0.875rem; text-transform: uppercase;
+             letter-spacing: 0.05em; margin-bottom: 0.75rem; }
+
+  .toc ol { padding-left: 1.25rem; }
+
+  .toc li { margin-bottom: 0.3rem; }
+
+  .toc a { color: #58a6ff; text-decoration: none; font-size: 0.9rem; }
+  .toc a:hover { text-decoration: underline; }
+
+  section {
+    margin-bottom: 4rem;
+    scroll-margin-top: 2rem;
+  }
+
+  section h2 {
+    font-size: 1.4rem;
+    color: #79c0ff;
+    margin-bottom: 0.4rem;
+  }
+
+  .subtitle {
+    color: #8b949e;
+    font-size: 0.875rem;
+    margin-bottom: 1.25rem;
+  }
+
+  .diagram-wrap {
+    background: #161b22;
+    border: 1px solid #30363d;
+    border-radius: 10px;
+    padding: 2rem;
+    overflow-x: auto;
+  }
+
+  .mermaid {
+    display: flex;
+    justify-content: center;
+  }
+
+  .legend {
+    margin-top: 1.25rem;
+    display: flex;
+    flex-wrap: wrap;
+    gap: 0.75rem;
+  }
+
+  .legend-item {
+    display: flex;
+    align-items: center;
+    gap: 0.4rem;
+    font-size: 0.8rem;
+    color: #8b949e;
+  }
+
+  .legend-dot {
+    width: 10px;
+    height: 10px;
+    border-radius: 50%;
+    flex-shrink: 0;
+  }
+
+  .note {
+    margin-top: 1rem;
+    padding: 0.75rem 1rem;
+    background: #1c2128;
+    border-left: 3px solid #388bfd;
+    border-radius: 0 6px 6px 0;
+    font-size: 0.85rem;
+    color: #8b949e;
+  }
+
+  .note strong { color: #c9d1d9; }
+
+  .grid-2 {
+    display: grid;
+    grid-template-columns: 1fr 1fr;
+    gap: 1.5rem;
+  }
+
+  @media (max-width: 900px) {
+    .grid-2 { grid-template-columns: 1fr; }
+    body { padding: 1rem; }
+  }
+</style>
+</head>
+<body>
+
+<header>
+  <h1>⚕️ Second Life — Architecture Diagrams</h1>
+  <p>DSCI 5260 · Group 7 · AI-powered Patient-to-Clinical-Trial Matching Platform · Flask + SQLite + Random Forest</p>
+</header>
+
+<div class="toc">
+  <h3>Contents</h3>
+  <ol>
+    <li><a href="#s1">System Architecture Overview</a></li>
+    <li><a href="#s2">Database Schema (ER Diagram)</a></li>
+    <li><a href="#s3">ML Pipeline — Boot, Train, Infer</a></li>
+    <li><a href="#s4">Patient User Journey (Sequence)</a></li>
+    <li><a href="#s5">Hospital User Journey (Sequence)</a></li>
+    <li><a href="#s6">Tiered Matching Logic</a></li>
+    <li><a href="#s7">Feature Engineering — 17 Features</a></li>
+  </ol>
+</div>
+
+<!-- =========================================================
+     1. SYSTEM ARCHITECTURE OVERVIEW
+     ========================================================= -->
+<section id="s1">
+  <h2>1. System Architecture Overview</h2>
+  <p class="subtitle">
+    Four-layer architecture: Browser &rarr; Flask Server &rarr; Business Logic (DB + ML) &rarr; Storage.
+    The pipeline boots in a background thread so the server is immediately available.
+  </p>
+  <div class="diagram-wrap">
+    <div class="mermaid">
+flowchart LR
+    subgraph BROWSER["🌐  Browser  (Client)"]
+        direction TB
+        LAND["<b>landing.html</b><br/>Login · Register<br/>(Patient &amp; Hospital)"]
+        subgraph PAT_SPA["Patient Portal — patient.html"]
+            direction TB
+            PA["My Profile<br/><small>conditions · meds · docs upload</small>"]
+            PB["Suggested Trials<br/><small>AI-ranked top-20</small>"]
+            PC["My Connections<br/><small>status · chat</small>"]
+            PD["Inbox 🔴<br/><small>unread count badge</small>"]
+        end
+        subgraph HOSP_SPA["Hospital Portal — hospital.html"]
+            direction TB
+            HA["Available Patients<br/><small>opt-in only, excl. connected</small>"]
+            HB["Search by Condition<br/><small>incl. connected toggle</small>"]
+            HC["My Trials<br/><small>active trials matched to hospital</small>"]
+            HD["Inbox 🔴<br/><small>unread count badge</small>"]
+            HE["My Connections<br/><small>accept · reject · complete · chat</small>"]
+            HF["My Profile<br/><small>name · location · conditions</small>"]
+        end
+    end
+
+    subgraph SERVER["⚙️  Flask Server — app.py  (:5000)"]
+        direction TB
+        subgraph AUTH_GRP["Auth (no session required)"]
+            AU["/auth/patient/register<br/>/auth/patient/login<br/>/auth/hospital/register<br/>/auth/hospital/login<br/>/auth/logout"]
+        end
+        subgraph PAT_GRP["Patient API"]
+            PAR["/api/patient/profile  GET · POST<br/>/api/patient/matches  GET<br/>/api/patient/interests  GET · POST · DELETE<br/>/api/patient/connections  GET<br/>/api/patient/connect  POST<br/>/api/patient/hospitals-for-trial  GET"]
+        end
+        subgraph MSG_GRP["Messaging + Inbox"]
+            MR["/api/patient/connections/&lt;cid&gt;/messages  GET · POST<br/>/api/patient/inbox  GET<br/>/api/hospital/connections/&lt;cid&gt;/messages  GET · POST<br/>/api/hospital/inbox  GET"]
+        end
+        subgraph HOSP_GRP["Hospital API"]
+            HAR["/api/hospital/profile  GET · POST<br/>/api/hospital/patients  GET<br/>/api/hospital/trials  GET<br/>/api/hospital/connect  POST<br/>/api/hospital/connections  GET<br/>/api/hospital/connections/&lt;cid&gt;/status  PUT"]
+        end
+        subgraph DOC_GRP["Documents"]
+            DR["/api/patient/documents  GET · POST<br/>/api/patient/documents/&lt;id&gt;  DELETE<br/>/api/patient/documents/&lt;id&gt;/download  GET"]
+        end
+        subgraph SHARED_GRP["Shared"]
+            SR["/api/status  GET<br/>/api/conditions/autocomplete  GET"]
+        end
+        BOOT["<b>Boot Thread</b><br/>init_db()<br/>pipeline.load()<br/>pipeline.train()"]
+    end
+
+    subgraph LOGIC["🧠  Business Logic"]
+        direction TB
+        subgraph DB_LAYER["database.py — SQLite Layer"]
+            T1[("patient_accounts<br/><small>25 rows</small>")]
+            T2[("hospital_accounts<br/><small>3 rows</small>")]
+            T3[("patient_trial_interests")]
+            T4[("connections")]
+            T5[("connection_messages")]
+        end
+        subgraph ML_LAYER["pipeline.py — ML Engine"]
+            ML1["Data Loader<br/><small>CSV → in-memory dicts</small>"]
+            ML2["Feature Engineer<br/><small>17 features per patient-trial pair</small>"]
+            ML3["RF + Isotonic Cal.<br/><small>90k pairs · AUC 0.60</small>"]
+            ML4["match_patient()"]
+            ML5["trials_for_hospital()"]
+            ML6["hospitals_for_trial()"]
+        end
+    end
+
+    subgraph STORAGE["💾  Storage"]
+        direction TB
+        SDB[("secondlife.db<br/><small>SQLite file</small>")]
+        SCACHE[("model_cache.pkl<br/><small>RF + calibrator</small>")]
+        SFILES["uploads/patient_docs/<br/><small>per-patient dirs</small>"]
+        SCSV_P["Synthea CSVs<br/><small>4 files · ~1M rows</small>"]
+        SCSV_T["ClinicalTrials CSVs<br/><small>7 files · ~2M rows</small>"]
+    end
+
+    LAND -->|"POST /auth/*"| AU
+    PAT_SPA -->|"REST/JSON"| PAR
+    PAT_SPA -->|"REST/JSON"| MSG_GRP
+    PAT_SPA -->|"multipart"| DOC_GRP
+    HOSP_SPA -->|"REST/JSON"| HAR
+    HOSP_SPA -->|"REST/JSON"| MSG_GRP
+    PAT_SPA & HOSP_SPA -->|"GET"| SHARED_GRP
+
+    AU & PAR & HAR --> DB_LAYER
+    MSG_GRP --> T4 & T5
+    DOC_GRP --> T1
+    DOC_GRP -->|"file I/O"| SFILES
+
+    PAR -->|"match_patient()"| ML4
+    PAR -->|"hospitals_for_trial()"| ML6
+    HAR -->|"trials_for_hospital()"| ML5
+
+    BOOT -->|"seeds data"| T1
+    BOOT -->|"load + train"| ML_LAYER
+
+    DB_LAYER <-->|"read/write"| SDB
+    ML3 <-->|"load/save"| SCACHE
+    ML1 -->|"read on boot"| SCSV_P
+    ML1 -->|"read on boot"| SCSV_T
+
+    ML1 --> ML2 --> ML3 --> ML4 & ML5 & ML6
+    </div>
+    <div class="note">
+      <strong>Boot sequence:</strong> On startup a background thread calls <code>init_db()</code> (seeds 25 patients + 3 hospitals),
+      then <code>pipeline.load()</code> (reads all CSVs into memory), then <code>pipeline.train()</code>
+      (builds RF model or loads <code>model_cache.pkl</code>). The Flask server is immediately available during boot;
+      <code>GET /api/status</code> returns <code>ready: false</code> until the thread completes.
+    </div>
+  </div>
+</section>
+
+<!-- =========================================================
+     2. DATABASE SCHEMA
+     ========================================================= -->
+<section id="s2">
+  <h2>2. Database Schema</h2>
+  <p class="subtitle">
+    SQLite database (<code>secondlife.db</code>) with 5 tables.
+    JSON columns (conditions, medications, documents, research_conditions) are stored as TEXT and parsed by
+    <code>_row_to_dict()</code> in database.py.
+  </p>
+  <div class="diagram-wrap">
+    <div class="mermaid">
+erDiagram
+    patient_accounts {
+        TEXT id PK
+        TEXT username UK
+        TEXT password_hash
+        TEXT synthea_id "null for hand-made patients"
+        TEXT first_name
+        TEXT last_name
+        TEXT dob
+        TEXT gender
+        TEXT address
+        TEXT conditions "JSON array of strings"
+        TEXT medications "JSON array of strings"
+        TEXT documents "JSON array of file metadata dicts"
+        INTEGER open_to_trials "0 or 1"
+        TEXT created_at
+    }
+
+    hospital_accounts {
+        TEXT id PK
+        TEXT username UK
+        TEXT password_hash
+        TEXT hospital_name
+        TEXT location "City, ST format"
+        TEXT research_conditions "JSON array of strings"
+        TEXT created_at
+    }
+
+    patient_trial_interests {
+        TEXT id PK
+        TEXT patient_id FK
+        TEXT trial_id "NCT number"
+        TEXT trial_title
+        REAL match_score
+        TEXT status "interested or withdrawn"
+        TEXT created_at
+    }
+
+    connections {
+        TEXT id PK
+        TEXT patient_id FK
+        TEXT hospital_id FK
+        TEXT trial_id "nullable NCT number"
+        TEXT trial_title
+        TEXT initiated_by "patient or hospital"
+        TEXT status "pending accepted rejected completed"
+        TEXT message
+        TEXT created_at
+    }
+
+    connection_messages {
+        TEXT id PK
+        TEXT connection_id FK
+        TEXT sender_role "patient or hospital"
+        TEXT sender_id
+        TEXT body "max 2000 chars"
+        TEXT created_at
+        INTEGER is_read "0 or 1"
+    }
+
+    patient_accounts ||--o{ patient_trial_interests : "saves interests"
+    patient_accounts ||--o{ connections : "initiates or receives"
+    hospital_accounts ||--o{ connections : "initiates or receives"
+    connections ||--o{ connection_messages : "contains messages"
+    </div>
+    <div class="note">
+      <strong>Uniqueness:</strong> <code>connections</code> has a composite unique index
+      <code>idx_conn_unique ON connections(patient_id, hospital_id, COALESCE(trial_id, ''))</code>
+      to prevent duplicate connection requests even when <code>trial_id</code> is NULL.
+      <code>patient_trial_interests</code> has <code>UNIQUE(patient_id, trial_id)</code>.
+      <code>connection_messages</code> has index <code>idx_msgs_conn ON (connection_id, created_at)</code>.
+    </div>
+  </div>
+</section>
+
+<!-- =========================================================
+     3. ML PIPELINE
+     ========================================================= -->
+<section id="s3">
+  <h2>3. ML Pipeline — Boot, Train, Infer</h2>
+  <p class="subtitle">
+    <code>pipeline.py</code> — Data loading from 11 CSV files, pseudo-label generation,
+    Random Forest training with isotonic calibration, and 3 inference functions.
+  </p>
+  <div class="diagram-wrap">
+    <div class="mermaid">
+flowchart TD
+    subgraph BOOT_TH["Boot Thread (background, server stays live during this)"]
+        direction LR
+        BT1["pipeline.load()"] --> BT2["pipeline.train()"]
+    end
+
+    subgraph LOAD_PHASE["load() — CSV Ingestion into Memory"]
+        direction LR
+        subgraph TRIAL_CSV["ClinicalTrials.gov CSVs"]
+            TC1["trail_studies.csv<br/><small>→ trial_profiles dict<br/>→ active_trial_ids set<br/>65k recruiting trials</small>"]
+            TC2["trail_conditions.csv<br/><small>→ trial_conditions dict<br/>34k trials with conditions</small>"]
+            TC3["trail_eligibilities.csv<br/><small>→ age range + gender per trial</small>"]
+            TC4["trail_facilities.csv<br/><small>→ trial_us_states dict<br/>→ trial_facility_tokens dict<br/>189k trials with US geo</small>"]
+            TC5["trail_interventions.csv<br/><small>→ trial_drug_keywords dict<br/>196k trials with drug data</small>"]
+            TC6["trail_brief_summaries.csv<br/><small>→ trial_summaries dict</small>"]
+            TC7["trail_keywords.csv<br/><small>→ trial_keywords dict</small>"]
+        end
+        subgraph SYNTHEA_CSV["Synthea Patient CSVs"]
+            SC1["patients_details.csv<br/><small>→ patient_demographics dict<br/>265k patients</small>"]
+            SC2["final_patients_conditions.csv<br/><small>→ patient_conditions dict<br/>106 overlapping conditions</small>"]
+            SC3["patients_medications.csv<br/><small>→ patient_medications dict<br/>213k patients</small>"]
+            SC4["patients_observations.csv<br/><small>→ patient_lab_types dict<br/>23k patients</small>"]
+        end
+    end
+
+    subgraph TRAIN_PHASE["train() — Model Training"]
+        direction TB
+        TR1["Sample 3000 patients × 30 trials<br/>+ random negatives<br/><small>GroupShuffleSplit 80/20 patient-level</small>"]
+        TR2["_compute_features()<br/><small>17-feature vector per patient-trial pair<br/>~90k training pairs total</small>"]
+        TR3["Pseudo-label generation<br/><small>6-feature weighted score >= 0.5<br/>+ 15% hash-deterministic noise<br/>→ 82.1% positive rate</small>"]
+        TR4["RandomForestClassifier<br/><small>n_estimators=200, max_depth=12<br/>class_weight=balanced</small>"]
+        TR5["CalibratedClassifierCV<br/><small>method=isotonic, cv=3</small>"]
+        TR6["Save model_cache.pkl<br/><small>skip training on next boot</small>"]
+        TR1 --> TR2 --> TR3 --> TR4 --> TR5 --> TR6
+    end
+
+    subgraph INFER["Inference — 3 Matching Functions"]
+        direction LR
+        IN1["<b>match_patient()</b><br/><small>Input: conditions, age, gender, patient_id<br/>Output: top-k trials ranked by<br/>combined_score = 0.6×eligibility_prob + 0.4×match_score</small>"]
+        IN2["<b>trials_for_hospital()</b><br/><small>Input: hospital name, location, research_conditions<br/>Output: active trials in 3 tiers<br/>Tier1: facility name Jaccard >= 0.25<br/>Tier2: same US state<br/>Tier3: condition overlap</small>"]
+        IN3["<b>hospitals_for_trial()</b><br/><small>Input: trial_id<br/>Output: DB hospitals in 4 tiers<br/>Tier1: name Jaccard >= 0.25<br/>Tier2: same state<br/>Tier3: condition overlap<br/>Tier4: fallback</small>"]
+    end
+
+    BOOT_TH --> LOAD_PHASE
+    BOOT_TH --> TRAIN_PHASE
+    LOAD_PHASE --> TR1
+    TR5 --> IN1
+    LOAD_PHASE --> IN2
+    LOAD_PHASE --> IN3
+    </div>
+    <div class="note">
+      <strong>Cache behaviour:</strong> If <code>model_cache.pkl</code> exists on disk, <code>train()</code> loads the cached RF + calibrator
+      and skips the ~2-3 min training step. Delete the file to force a full retrain (required after changing features or pipeline logic).
+    </div>
+  </div>
+</section>
+
+<!-- =========================================================
+     4. PATIENT JOURNEY
+     ========================================================= -->
+<section id="s4">
+  <h2>4. Patient User Journey</h2>
+  <p class="subtitle">
+    Key request-response flows through the patient portal, from registration to messaging.
+  </p>
+  <div class="diagram-wrap">
+    <div class="mermaid">
+sequenceDiagram
+    actor Patient
+    participant UI as patient.html
+    participant Flask as Flask (app.py)
+    participant DB as database.py
+    participant ML as pipeline.py
+    participant FS as File System
+
+    Note over Patient,FS: Registration / Login
+    Patient->>UI: Fill register form
+    UI->>Flask: POST /auth/patient/register
+    Flask->>DB: create_patient()
+    DB-->>Flask: patient dict (or 409 username taken)
+    Flask-->>UI: session cookie + patient JSON
+    UI->>UI: Render portal, call checkStatus()
+
+    Note over Patient,FS: Pipeline ready check
+    UI->>Flask: GET /api/status
+    Flask-->>UI: ready: false (202) or ready: true + stats
+    UI->>UI: Show yellow banner if not ready; startStatusPoll() every 5s
+
+    Note over Patient,FS: Profile update
+    Patient->>UI: Edit conditions / medications / toggle open_to_trials
+    UI->>Flask: POST /api/patient/profile
+    Flask->>DB: update_patient_profile()
+
+    Note over Patient,FS: Get AI trial matches
+    Patient->>UI: Click "Suggested Trials" tab
+    UI->>Flask: GET /api/patient/matches
+    Flask->>ML: match_patient(conditions, age, gender, patient_id)
+    ML->>ML: _compute_features() x all active trials
+    ML->>ML: RF.predict_proba() → calibrated eligibility prob
+    ML->>ML: combined_score = 0.6×prob + 0.4×match_score
+    ML-->>Flask: top-20 trials sorted by combined_score
+    Flask-->>UI: results array with 24 fields per trial
+
+    Note over Patient,FS: Connect with a hospital
+    Patient->>UI: Click "Connect with Hospital" on a trial card
+    UI->>Flask: GET /api/patient/hospitals-for-trial?trial_id=NCT...
+    Flask->>ML: hospitals_for_trial(trial_id)
+    ML-->>Flask: tiered hospital list (Tier 1..4)
+    Flask-->>UI: hospitals array
+    UI->>UI: Render modal — green section (Tier1) + grey section (Tier2-4)
+    Patient->>UI: Select hospital + enter message + submit
+    UI->>Flask: POST /api/patient/connect
+    Flask->>DB: connection_exists()? if no → create_connection()
+    DB-->>Flask: connection dict (or 409 duplicate)
+    Flask-->>UI: success redirect to connections tab
+
+    Note over Patient,FS: Inbox and messaging
+    Patient->>UI: Click "Inbox" tab
+    UI->>Flask: GET /api/patient/inbox
+    Flask->>DB: get_patient_inbox_threads(patient_id)
+    DB-->>Flask: threads with last_message + unread_count
+    Flask-->>UI: threads array + update badge
+    Patient->>UI: Click "Open" on a thread
+    UI->>Flask: GET /api/patient/connections/{cid}/messages
+    Flask->>DB: mark_messages_read(role=hospital)
+    Flask->>DB: get_connection_messages(cid)
+    Flask-->>UI: messages array
+    Patient->>UI: Type reply + Enter or Send
+    UI->>Flask: POST /api/patient/connections/{cid}/messages
+    Flask->>DB: create_connection_message(role=patient)
+
+    Note over Patient,FS: Document upload
+    Patient->>UI: Select file in Documents section
+    UI->>Flask: POST /api/patient/documents (multipart/form-data)
+    Flask->>Flask: secure_filename, validate extension + 10MB limit
+    Flask->>FS: Save to uploads/patient_docs/{pid}/{doc_id}_filename
+    Flask->>DB: update_patient_profile(documents=[...append])
+    Flask-->>UI: document metadata dict
+    </div>
+  </div>
+</section>
+
+<!-- =========================================================
+     5. HOSPITAL JOURNEY
+     ========================================================= -->
+<section id="s5">
+  <h2>5. Hospital User Journey</h2>
+  <p class="subtitle">
+    Key request-response flows through the hospital portal, from login to connection management and messaging.
+  </p>
+  <div class="diagram-wrap">
+    <div class="mermaid">
+sequenceDiagram
+    actor Hospital
+    participant UI as hospital.html
+    participant Flask as Flask (app.py)
+    participant DB as database.py
+    participant ML as pipeline.py
+
+    Note over Hospital,ML: Login
+    Hospital->>UI: Enter credentials
+    UI->>Flask: POST /auth/hospital/login
+    Flask->>DB: authenticate_hospital()
+    DB-->>Flask: hospital dict
+    Flask-->>UI: session cookie + hospital JSON
+
+    Note over Hospital,ML: Browse available patients
+    Hospital->>UI: Click "Available Patients" tab (default)
+    UI->>Flask: GET /api/hospital/patients
+    Flask->>DB: get_open_patients_for_hospital(include_connected=False)
+    DB-->>Flask: patients with open_to_trials=1 excluding already-connected
+    Flask-->>UI: patient cards (25 opt-in patients total)
+
+    Note over Hospital,ML: Search by condition
+    Hospital->>UI: Type condition + toggle "include connected"
+    UI->>Flask: GET /api/hospital/patients?condition=diabetes&include_connected=true
+    Flask->>DB: get_open_patients_for_hospital(condition_filter, include_connected=True)
+    Flask-->>UI: filtered patient list
+
+    Note over Hospital,ML: View matched trials
+    Hospital->>UI: Click "My Trials" tab
+    UI->>Flask: GET /api/hospital/trials
+    Flask->>ML: trials_for_hospital(hospital_name, location, research_conditions)
+    ML->>ML: Tier1 — facility name Jaccard >= 0.25 + is_active
+    ML->>ML: Tier2 — same US state + is_active
+    ML->>ML: Tier3 — condition overlap + is_active
+    ML-->>Flask: active trial list with match_tier and match_reason
+    Flask-->>UI: trial cards with status, phase, facility, ClinicalTrials.gov link
+
+    Note over Hospital,ML: Connect with a patient
+    Hospital->>UI: Click "Connect" on a patient card
+    UI->>Flask: POST /api/hospital/connect
+    Flask->>DB: create_connection(initiated_by=hospital)
+    DB-->>Flask: connection dict (or 409 duplicate)
+
+    Note over Hospital,ML: Manage connections
+    Hospital->>UI: Click "My Connections" tab
+    UI->>Flask: GET /api/hospital/connections
+    Flask->>DB: get_hospital_connections(hospital_id)
+    Flask-->>UI: connection table with patient details
+    Hospital->>UI: Click "Accept" / "Reject" / "Complete"
+    UI->>Flask: PUT /api/hospital/connections/{cid}/status
+    Flask->>DB: update_connection_status(cid, status)
+
+    Note over Hospital,ML: Inbox and messaging
+    Hospital->>UI: Click "Inbox" tab
+    UI->>Flask: GET /api/hospital/inbox
+    Flask->>DB: get_hospital_inbox_threads(hospital_id)
+    DB-->>Flask: threads sorted by last_message_at DESC
+    Flask-->>UI: thread list with unread_count, update nav badge
+    Hospital->>UI: Click "Open" on a thread
+    UI->>Flask: GET /api/hospital/connections/{cid}/messages
+    Flask->>DB: mark_messages_read(role=hospital)
+    Flask->>DB: get_connection_messages(cid)
+    Flask-->>UI: messages in chronological order
+    Hospital->>UI: Type reply + Send
+    UI->>Flask: POST /api/hospital/connections/{cid}/messages
+    Flask->>DB: create_connection_message(role=hospital)
+
+    Note over Hospital,ML: Update profile
+    Hospital->>UI: Edit name / location / research conditions
+    UI->>Flask: POST /api/hospital/profile
+    Flask->>DB: update_hospital_profile()
+    Flask-->>UI: updated hospital dict
+    UI->>UI: Update navbar name live (no page reload)
+    </div>
+  </div>
+</section>
+
+<!-- =========================================================
+     6. TIERED MATCHING
+     ========================================================= -->
+<section id="s6">
+  <h2>6. Tiered Matching Logic</h2>
+  <p class="subtitle">
+    Two reverse-matching functions sharing the same 4-tier cascade. Both use
+    <code>_name_tokens()</code> + Jaccard similarity for Tier 1, US state extraction for Tier 2,
+    and condition set intersection for Tier 3.
+  </p>
+  <div class="grid-2">
+    <div>
+      <h3 style="color:#3fb950;margin-bottom:.75rem;font-size:1rem;">
+        hospitals_for_trial(trial_id) — Patient selects a trial
+      </h3>
+      <div class="diagram-wrap">
+        <div class="mermaid">
+flowchart TD
+    START_A["Patient clicks<br/>'Connect with Hospital'<br/>on trial NCTxxxxxx"]
+
+    subgraph T1A["Tier 1 — Verified Site"]
+        C1A["For each hospital in DB:<br/>Jaccard(hospital name tokens,<br/>ANY trial facility name tokens) >= 0.25"]
+        R1A["🟢 match_reason:<br/>'verified site on this trial'"]
+    end
+
+    subgraph T2A["Tier 2 — Same State"]
+        C2A["Hospital state extracted from<br/>'City, ST' location field<br/>== any trial US facility state"]
+        R2A["⬜ match_reason:<br/>'in same state as a trial site'"]
+    end
+
+    subgraph T3A["Tier 3 — Condition Overlap"]
+        C3A["hospital.research_conditions<br/>∩ trial.conditions ≠ empty"]
+        R3A["⬜ match_reason:<br/>'researches related conditions'"]
+    end
+
+    subgraph T4A["Tier 4 — Fallback"]
+        C4A["Trial has no US facility data<br/>all remaining hospitals included"]
+        R4A["⬜ match_reason: ''"]
+    end
+
+    MODAL["Render connect modal<br/>🟢 VERIFIED TRIAL SITES header<br/>⬜ RELATED HOSPITALS header<br/>(Tiers 2–4)"]
+
+    START_A --> T1A --> T2A --> T3A --> T4A --> MODAL
+        </div>
+      </div>
+    </div>
+
+    <div>
+      <h3 style="color:#79c0ff;margin-bottom:.75rem;font-size:1rem;">
+        trials_for_hospital(name, loc, conditions) — Hospital views My Trials
+      </h3>
+      <div class="diagram-wrap">
+        <div class="mermaid">
+flowchart TD
+    START_B["Hospital opens<br/>'My Trials' tab"]
+
+    subgraph T1B["Tier 1 — Name Matched"]
+        C1B["For each active trial:<br/>Jaccard(hospital name tokens,<br/>trial facility name tokens) >= 0.25<br/>+ is_active = True"]
+        R1B["match_tier: 1<br/>match_reason:<br/>'name matched to trial site'"]
+    end
+
+    subgraph T2B["Tier 2 — Same State"]
+        C2B["Hospital state<br/>== trial US facility state<br/>+ is_active = True"]
+        R2B["match_tier: 2<br/>match_reason:<br/>'in same state as trial site'"]
+    end
+
+    subgraph T3B["Tier 3 — Condition Overlap"]
+        C3B["hospital.research_conditions<br/>∩ trial.conditions ≠ empty<br/>+ is_active = True"]
+        R3B["match_tier: 3<br/>match_reason:<br/>'researches related conditions'"]
+    end
+
+    SORT["Sort by match_tier ASC<br/>cap at top_k=20<br/>Return trial cards with<br/>status · phase · facility · summary"]
+
+    START_B --> T1B --> T2B --> T3B --> SORT
+        </div>
+      </div>
+    </div>
+  </div>
+
+  <div style="margin-top:1.5rem">
+    <h3 style="color:#c9d1d9;margin-bottom:.75rem;font-size:1rem;">Shared Helpers</h3>
+    <div class="diagram-wrap">
+      <div class="mermaid">
+flowchart LR
+    subgraph TOK["_name_tokens(name)"]
+        TK1["Lowercase all words"]
+        TK2["Remove stopwords:<br/>hospital, medical, center, clinic, university,<br/>health, care, system, institute, general,<br/>regional, national, community, the, of, and, at…"]
+        TK3["Remove words shorter than 3 chars"]
+        TK4["Return frozenset of tokens"]
+        TK1 --> TK2 --> TK3 --> TK4
+    end
+
+    subgraph JAC["Jaccard Similarity"]
+        J1["score = |A ∩ B| / |A ∪ B|"]
+        J2["Best score across ALL<br/>facility names for the trial"]
+        J1 --> J2
+    end
+
+    subgraph STATE["State Extraction"]
+        S1["Parse 'City, ST' location string"]
+        S2["Lookup ST abbreviation<br/>in US_STATE_MAP dict<br/>→ full name e.g. 'Massachusetts'"]
+        S3["Compare against<br/>trial_us_states set<br/>(from trail_facilities.csv)"]
+        S1 --> S2 --> S3
+    end
+
+    TOK --> JAC
+    TOK --> STATE
+      </div>
+    </div>
+  </div>
+</section>
+
+<!-- =========================================================
+     7. FEATURE ENGINEERING
+     ========================================================= -->
+<section id="s7">
+  <h2>7. Feature Engineering — 17 Features</h2>
+  <p class="subtitle">
+    All features are normalised to 0–1. Computed by <code>_compute_features(patient_row, trial_id)</code>
+    in pipeline.py. Used for both training pseudo-labels and inference scoring.
+  </p>
+  <div class="diagram-wrap">
+    <div class="mermaid">
+flowchart LR
+    INPUT["Patient Profile<br/>+ Trial Profile"]
+
+    subgraph COND_GRP["Condition Overlap (4)"]
+        direction TB
+        CF1["condition_overlap<br/><small>raw count: |patient_conds ∩ trial_conds|</small>"]
+        CF2["jaccard_similarity<br/><small>overlap / union</small>"]
+        CF3["overlap_ratio_trial<br/><small>overlap / |trial_conds|</small>"]
+        CF4["overlap_ratio_patient<br/><small>overlap / |patient_conds|</small>"]
+    end
+
+    subgraph PROF_GRP["Condition Profile (5)"]
+        direction TB
+        PF1["condition_rarity_score<br/><small>mean(1/log2(n_trials_per_cond + 2)), norm 0-1</small>"]
+        PF2["trial_specificity<br/><small>1 / |trial_conds|</small>"]
+        PF3["condition_burden<br/><small>total_patient_conds / 10</small>"]
+        PF4["active_ratio<br/><small>active_conds / total_conds</small>"]
+        PF5["resolved_ratio<br/><small>resolved_conds / total_conds</small>"]
+    end
+
+    subgraph ELIG_GRP["Eligibility (4)"]
+        direction TB
+        EF1["age_distance<br/><small>norm distance outside [min_age, max_age]</small>"]
+        EF2["age_centered<br/><small>position within range, -1 to +1</small>"]
+        EF3["age_compatibility<br/><small>1.0 in range; decays over 30yr gap</small>"]
+        EF4["gender_compatibility<br/><small>1.0 match/ALL; 0.1 mismatch</small>"]
+    end
+
+    subgraph EXT_GRP["External Signals (3)"]
+        direction TB
+        XF1["geo_feasibility<br/><small>1.0 same state · 0.75 other US state · 0.5 no data</small>"]
+        XF2["med_compatibility<br/><small>patient med keywords ∩ trial drug intervention keywords</small>"]
+        XF3["lab_availability<br/><small>patient observation types covered by trial / 20</small>"]
+    end
+
+    subgraph META_GRP["Metadata (1)"]
+        MF1["data_completeness<br/><small>fraction of key fields non-empty<br/>importance: ~0.0 in trained RF</small>"]
+    end
+
+    RF["RandomForest<br/>predict_proba()<br/>→ eligibility_prob"]
+
+    INPUT --> COND_GRP
+    INPUT --> PROF_GRP
+    INPUT --> ELIG_GRP
+    INPUT --> EXT_GRP
+    INPUT --> META_GRP
+
+    COND_GRP --> RF
+    PROF_GRP --> RF
+    ELIG_GRP --> RF
+    EXT_GRP --> RF
+    META_GRP --> RF
+    </div>
+
+    <div class="note" style="margin-top:1rem">
+      <strong>Top features by importance (Random Forest):</strong>
+      age_distance (20.4%) · age_compatibility (16.9%) · gender_compatibility (12.9%) · age_centered (7.9%) ·
+      jaccard_similarity (7.5%) · condition_rarity_score (6.2%) · overlap_ratio_trial (5.2%) · overlap_ratio_patient (5.1%) ·
+      condition_overlap (4.0%) · remaining 8 features (&lt;3% each).
+      <br/><br/>
+      <strong>Note on AUC (0.60):</strong> The 82.1% positive pseudo-label rate makes the classification trivially high-accuracy
+      but limits discriminative power. Raising the label threshold from 0.5 → 0.6 or enforcing equal pos/neg sampling would improve AUC.
+    </div>
+  </div>
+</section>
+
+<footer style="border-top:1px solid #30363d;margin-top:3rem;padding-top:1.5rem;color:#8b949e;font-size:0.8rem;text-align:center">
+  Second Life · DSCI 5260 · Group 7 · Generated 2026-04-25
+</footer>
+
+<script>
+  mermaid.initialize({
+    startOnLoad: true,
+    theme: 'dark',
+    themeVariables: {
+      primaryColor:        '#1f6feb',
+      primaryTextColor:    '#c9d1d9',
+      primaryBorderColor:  '#388bfd',
+      lineColor:           '#58a6ff',
+      secondaryColor:      '#161b22',
+      tertiaryColor:       '#1c2128',
+      background:          '#0d1117',
+      mainBkg:             '#161b22',
+      nodeBorder:          '#30363d',
+      clusterBkg:          '#13161b',
+      clusterBorder:       '#30363d',
+      titleColor:          '#79c0ff',
+      edgeLabelBackground: '#161b22',
+      fontFamily:          'system-ui, sans-serif',
+    },
+    flowchart: {
+      useMaxWidth: false,
+      htmlLabels:  true,
+      curve:       'basis',
+      padding:     20,
+    },
+    sequence: {
+      useMaxWidth:    false,
+      mirrorActors:   false,
+      messageAlign:   'center',
+      actorFontFamily:'system-ui, sans-serif',
+      noteFontFamily: 'system-ui, sans-serif',
+    },
+    er: {
+      useMaxWidth: false,
+    }
+  });
+</script>
+</body>
+</html>

Dockerfile

@@ -0,0 +1,16 @@
+FROM python:3.11-slim
+
+ENV PYTHONUNBUFFERED=1
+ENV PORT=7860
+ENV HF_DATASET_REPO=MrNoOne07/second-life-data
+
+WORKDIR /app
+
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+
+EXPOSE 7860
+
+CMD ["python", "hf_space_bootstrap.py"]

README.md

@@ -1,10 +1,32 @@
 ---
-title: Second Life
-emoji: 🔥
-colorFrom: pink
-colorTo: red
+title: Second Life Clinical Trial Matching
 sdk: docker
+app_port: 7860
 pinned: false
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# Second Life Clinical Trial Matching
+
+Second Life is a two-portal Flask application for matching synthetic patient
+profiles to clinical trials and helping hospitals identify patients who are open
+to trial participation.
+
+## Space Startup
+
+This Hugging Face Space uses Docker. On startup, `hf_space_bootstrap.py`
+downloads the required dataset zip files from the companion Dataset repository:
+
+`MrNoOne07/second-life-data`
+
+The app then starts `app.py` on port `7860`.
+
+## Demo Logins
+
+Demo credentials are documented in `login.md`.
+
+## Data Notice
+
+The uploaded demo data includes generated Synthea patient data and processed
+ClinicalTrials.gov data for class demonstration. MIMIC-IV validation data is not
+bundled in this Space by default because it has separate access and licensing
+requirements.

app.py

@@ -0,0 +1,760 @@
+"""
+Second Life — Flask Web Application
+DSCI 5260 | Group 7
+
+Two-portal system: Patient Portal + Hospital Portal
+Run:  python app.py   →  http://localhost:5000
+"""
+
+import os
+import sys
+import uuid
+import mimetypes
+import secrets
+import threading
+from datetime import datetime
+from pathlib import Path
+
+from flask import (Flask, jsonify, redirect, render_template,
+                   request, send_file, session, url_for)
+from werkzeug.utils import secure_filename
+
+sys.path.insert(0, str(Path(__file__).parent))
+from pipeline import SecondLifePipeline
+import database as db
+
+app = Flask(__name__)
+app.secret_key = os.environ.get("SECRET_KEY", secrets.token_hex(32))
+app.config["JSON_SORT_KEYS"] = False
+app.config["MAX_CONTENT_LENGTH"] = 10 * 1024 * 1024  # 10 MB upload limit
+app.config["TEMPLATES_AUTO_RELOAD"] = True
+
+UPLOAD_DIR = Path(__file__).parent / "uploads" / "patient_docs"
+ALLOWED_EXTENSIONS = {".pdf", ".docx", ".doc", ".txt", ".png", ".jpg", ".jpeg"}
+
+# ---------------------------------------------------------------------------
+# Pipeline boot (background thread — server comes up immediately)
+# ---------------------------------------------------------------------------
+pipeline: SecondLifePipeline | None = None
+_boot_error: str = ""
+_boot_ready = threading.Event()
+
+
+def _boot():
+    global pipeline, _boot_error
+    try:
+        db.init_db()
+        p = SecondLifePipeline()
+        p.load()
+        p.train()
+        pipeline = p
+        print("[app] Pipeline ready.")
+    except Exception as e:
+        import traceback
+        _boot_error = traceback.format_exc()
+        print(f"[app] BOOT ERROR:\n{_boot_error}", file=sys.stderr)
+    finally:
+        _boot_ready.set()
+
+
+_t = threading.Thread(target=_boot, daemon=True)
+_t.start()
+
+
+def _ready():
+    return pipeline is not None
+
+
+# ---------------------------------------------------------------------------
+# Auth guards (return (response, code) or None)
+# ---------------------------------------------------------------------------
+
+def _patient_required():
+    if "patient_id" not in session:
+        return jsonify({"error": "Not authenticated"}), 401
+    return None
+
+
+def _hospital_required():
+    if "hospital_id" not in session:
+        return jsonify({"error": "Not authenticated"}), 401
+    return None
+
+
+# ---------------------------------------------------------------------------
+# Sanitisers (strip password hash before sending to client)
+# ---------------------------------------------------------------------------
+
+def _pub_patient(p: dict) -> dict:
+    d = dict(p)
+    d.pop("password_hash", None)
+    return d
+
+
+def _pub_hospital(h: dict) -> dict:
+    d = dict(h)
+    d.pop("password_hash", None)
+    return d
+
+
+# ---------------------------------------------------------------------------
+# Page routes
+# ---------------------------------------------------------------------------
+
+@app.route("/")
+def landing():
+    if "patient_id" in session:
+        return redirect(url_for("patient_portal"))
+    if "hospital_id" in session:
+        return redirect(url_for("hospital_portal"))
+    return render_template("landing.html")
+
+
+@app.route("/patient")
+def patient_portal():
+    if "patient_id" not in session:
+        return redirect(url_for("landing"))
+    patient = db.get_patient_by_id(session["patient_id"])
+    if not patient:
+        session.clear()
+        return redirect(url_for("landing"))
+    return render_template("patient.html", patient=_pub_patient(patient))
+
+
+@app.route("/hospital")
+def hospital_portal():
+    if "hospital_id" not in session:
+        return redirect(url_for("landing"))
+    hospital = db.get_hospital_by_id(session["hospital_id"])
+    if not hospital:
+        session.clear()
+        return redirect(url_for("landing"))
+    return render_template("hospital.html", hospital=_pub_hospital(hospital))
+
+
+# ---------------------------------------------------------------------------
+# Auth routes
+# ---------------------------------------------------------------------------
+
+@app.route("/auth/patient/register", methods=["POST"])
+def patient_register():
+    data = request.get_json(force=True) or {}
+    username   = data.get("username",   "").strip()
+    password   = data.get("password",   "").strip()
+    first_name = data.get("first_name", "").strip()
+    last_name  = data.get("last_name",  "").strip()
+    dob        = data.get("dob",        "").strip()
+    gender     = data.get("gender",     "").strip().upper()
+    address    = data.get("address",    "").strip()
+
+    if not username or not password:
+        return jsonify({"error": "Username and password are required"}), 400
+
+    patient = db.create_patient(username, password, first_name, last_name,
+                                dob, gender, address)
+    if patient is None:
+        return jsonify({"error": "Username already taken"}), 409
+
+    session["patient_id"] = patient["id"]
+    session["role"] = "patient"
+    return jsonify({"success": True, "patient": _pub_patient(patient)})
+
+
+@app.route("/auth/patient/login", methods=["POST"])
+def patient_login():
+    data    = request.get_json(force=True) or {}
+    patient = db.authenticate_patient(data.get("username", ""),
+                                      data.get("password", ""))
+    if patient is None:
+        return jsonify({"error": "Invalid username or password"}), 401
+
+    session["patient_id"] = patient["id"]
+    session["role"] = "patient"
+    return jsonify({"success": True, "patient": _pub_patient(patient)})
+
+
+@app.route("/auth/hospital/register", methods=["POST"])
+def hospital_register():
+    data          = request.get_json(force=True) or {}
+    username      = data.get("username",      "").strip()
+    password      = data.get("password",      "").strip()
+    hospital_name = data.get("hospital_name", "").strip()
+    location      = data.get("location",      "").strip()
+    research_conditions = data.get("research_conditions", [])
+
+    if not username or not password or not hospital_name:
+        return jsonify({"error": "Username, password and hospital name are required"}), 400
+
+    hospital = db.create_hospital(username, password, hospital_name,
+                                  location, research_conditions)
+    if hospital is None:
+        return jsonify({"error": "Username already taken"}), 409
+
+    session["hospital_id"] = hospital["id"]
+    session["role"] = "hospital"
+    return jsonify({"success": True, "hospital": _pub_hospital(hospital)})
+
+
+@app.route("/auth/hospital/login", methods=["POST"])
+def hospital_login():
+    data     = request.get_json(force=True) or {}
+    hospital = db.authenticate_hospital(data.get("username", ""),
+                                        data.get("password", ""))
+    if hospital is None:
+        return jsonify({"error": "Invalid username or password"}), 401
+
+    session["hospital_id"] = hospital["id"]
+    session["role"] = "hospital"
+    return jsonify({"success": True, "hospital": _pub_hospital(hospital)})
+
+
+@app.route("/auth/logout", methods=["POST"])
+def logout():
+    session.clear()
+    return jsonify({"success": True})
+
+
+# ---------------------------------------------------------------------------
+# Patient API
+# ---------------------------------------------------------------------------
+
+@app.route("/api/patient/profile", methods=["GET"])
+def get_patient_profile():
+    err = _patient_required()
+    if err:
+        return err
+    patient = db.get_patient_by_id(session["patient_id"])
+    if not patient:
+        return jsonify({"error": "Not found"}), 404
+    return jsonify(_pub_patient(patient))
+
+
+@app.route("/api/patient/profile", methods=["POST"])
+def update_patient_profile():
+    err = _patient_required()
+    if err:
+        return err
+    data    = request.get_json(force=True) or {}
+    allowed = {"first_name", "last_name", "dob", "gender", "address",
+               "conditions", "medications", "open_to_trials"}
+    kwargs  = {k: v for k, v in data.items() if k in allowed}
+    db.update_patient_profile(session["patient_id"], **kwargs)
+    return jsonify(_pub_patient(db.get_patient_by_id(session["patient_id"])))
+
+
+@app.route("/api/patient/matches", methods=["GET"])
+def patient_matches():
+    err = _patient_required()
+    if err:
+        return err
+    if not _ready():
+        return jsonify({"error": "System still loading. Please wait."}), 503
+
+    patient    = db.get_patient_by_id(session["patient_id"])
+    conditions = patient.get("conditions", [])
+    address    = patient.get("address", "")
+
+    # Calculate age from DOB
+    age  = 50
+    dob  = patient.get("dob", "")
+    if dob:
+        try:
+            birth = datetime.strptime(dob, "%Y-%m-%d")
+            age   = int((datetime.now() - birth).days / 365.25)
+        except Exception:
+            pass
+
+    gender = patient.get("gender", "M") or "M"
+
+    if not conditions:
+        return jsonify({"results": [],
+                        "message": "No conditions on your profile. "
+                                   "Update your profile first."})
+
+    matches = pipeline.match_patient(
+        conditions, age, gender, top_k=20,
+        patient_id=session["patient_id"], address=address,
+    )
+
+    # Annotate with saved interest status
+    interests = {i["trial_id"]: i["status"]
+                 for i in db.get_patient_interests(session["patient_id"])}
+    for m in matches:
+        m["interest_status"] = interests.get(m["trial_id"])
+
+    return jsonify({"results": matches, "total": len(matches)})
+
+
+@app.route("/api/patient/interests", methods=["GET"])
+def get_patient_interests():
+    err = _patient_required()
+    if err:
+        return err
+    return jsonify({"interests": db.get_patient_interests(session["patient_id"])})
+
+
+@app.route("/api/patient/interest", methods=["POST"])
+def save_patient_interest():
+    err = _patient_required()
+    if err:
+        return err
+    data       = request.get_json(force=True) or {}
+    trial_id   = data.get("trial_id", "")
+    trial_title = data.get("trial_title", "")
+    match_score = float(data.get("match_score", 0))
+    if not trial_id:
+        return jsonify({"error": "trial_id required"}), 400
+    db.save_trial_interest(session["patient_id"], trial_id,
+                           trial_title, match_score)
+    return jsonify({"success": True})
+
+
+@app.route("/api/patient/interest/<trial_id>", methods=["DELETE"])
+def remove_patient_interest(trial_id):
+    err = _patient_required()
+    if err:
+        return err
+    db.withdraw_interest(session["patient_id"], trial_id)
+    return jsonify({"success": True})
+
+
+@app.route("/api/patient/connections", methods=["GET"])
+def get_patient_connections():
+    err = _patient_required()
+    if err:
+        return err
+    return jsonify({"connections": db.get_patient_connections(session["patient_id"])})
+
+
+@app.route("/api/patient/connect", methods=["POST"])
+def patient_connect():
+    err = _patient_required()
+    if err:
+        return err
+    data        = request.get_json(force=True) or {}
+    hospital_id = data.get("hospital_id", "")
+    trial_id    = data.get("trial_id", "")
+    trial_title = data.get("trial_title", "")
+    message     = data.get("message", "")
+    if not hospital_id:
+        return jsonify({"error": "hospital_id required"}), 400
+    conn = db.create_connection(
+        session["patient_id"], hospital_id, trial_id, trial_title,
+        initiated_by="patient", message=message,
+    )
+    if conn is None:
+        return jsonify({"error": "A connection with this hospital for this trial already exists"}), 409
+    return jsonify({"success": True, "connection": conn})
+
+
+def _hospital_name_tokens(name: str) -> frozenset:
+    """Tokenise a hospital name for facility-name matching (same stopword set as pipeline)."""
+    import re as _re
+    _STOP = {"the","of","and","at","for","in","a","an","is","by",
+             "hospital","medical","center","centre","clinic","university",
+             "health","care","healthcare","system","institute","foundation",
+             "research","general","regional","national","community",
+             "services","department","division","college","school"}
+    words = _re.findall(r"[a-z]+", str(name).lower())
+    return frozenset(w for w in words if w not in _STOP and len(w) >= 3)
+
+
+def _facility_match_score(h_tokens: frozenset, facility_token_list: list) -> float:
+    """
+    Return the best Jaccard score between hospital name tokens and any facility
+    name tokens for a given trial.  Returns 0.0 if facility_token_list is empty.
+    """
+    if not h_tokens or not facility_token_list:
+        return 0.0
+    best = 0.0
+    for fac_tokens in facility_token_list:
+        if not fac_tokens:
+            continue
+        inter = len(h_tokens & fac_tokens)
+        union = len(h_tokens | fac_tokens)
+        score = inter / union if union else 0.0
+        if score > best:
+            best = score
+    return best
+
+
+@app.route("/api/patient/hospitals-for-trial", methods=["GET"])
+def hospitals_for_trial():
+    err = _patient_required()
+    if err:
+        return err
+    trial_id = request.args.get("trial_id", "")
+    if not trial_id:
+        return jsonify({"hospitals": []})
+
+    from pipeline import STATE_ABBREV
+    import re as _re
+
+    trial_fac_tokens = []
+    trial_states     = set()
+    trial_conds      = set()
+    if _ready():
+        trial_fac_tokens = pipeline.trial_facility_tokens.get(trial_id, [])
+        trial_states     = pipeline.trial_us_states.get(trial_id, set())
+        if trial_id in pipeline.trial_profiles:
+            trial_conds = pipeline.trial_profiles[trial_id]["conditions"]
+
+    all_hospitals = db.get_all_hospitals()
+    result = []
+    for h in all_hospitals:
+        h_name = (h.get("hospital_name") or "").strip()
+        h_loc  = (h.get("location") or "").strip()
+
+        # Extract hospital state from "City, ST" location string
+        m = _re.search(r",\s*([A-Z]{2})\s*$", h_loc)
+        h_state_full = STATE_ABBREV.get(m.group(1), "") if m else ""
+
+        # Tier 1: real facility-name match (Jaccard ≥ 0.25 is a generous but meaningful threshold)
+        h_tokens     = _hospital_name_tokens(h_name)
+        name_score   = _facility_match_score(h_tokens, trial_fac_tokens)
+        if name_score >= 0.25:
+            result.append({
+                "id":            h["id"],
+                "hospital_name": h_name,
+                "location":      h_loc,
+                "match_reason":  "verified site on this trial",
+                "match_tier":    1,
+            })
+            continue
+
+        # Tier 2: hospital is in the same US state as a trial facility
+        if trial_states and h_state_full and h_state_full in trial_states:
+            result.append({
+                "id":            h["id"],
+                "hospital_name": h_name,
+                "location":      h_loc,
+                "match_reason":  "in same state as a trial site",
+                "match_tier":    2,
+            })
+            continue
+
+        # Tier 3: research_conditions overlap with trial conditions
+        rc_set = {r.lower() for r in (h.get("research_conditions") or [])}
+        if rc_set & trial_conds:
+            result.append({
+                "id":            h["id"],
+                "hospital_name": h_name,
+                "location":      h_loc,
+                "match_reason":  "researches related conditions",
+                "match_tier":    3,
+            })
+            continue
+
+        # Tier 4: no facility data exists at all — show all registered hospitals
+        if not trial_fac_tokens and not trial_states:
+            result.append({
+                "id":            h["id"],
+                "hospital_name": h_name,
+                "location":      h_loc,
+                "match_reason":  "",
+                "match_tier":    4,
+            })
+
+    result.sort(key=lambda x: x["match_tier"])
+    return jsonify({"hospitals": result})
+
+
+# ---------------------------------------------------------------------------
+# Hospital API
+# ---------------------------------------------------------------------------
+
+@app.route("/api/hospital/profile", methods=["GET"])
+def get_hospital_profile():
+    err = _hospital_required()
+    if err:
+        return err
+    hospital = db.get_hospital_by_id(session["hospital_id"])
+    if not hospital:
+        return jsonify({"error": "Not found"}), 404
+    return jsonify(_pub_hospital(hospital))
+
+
+@app.route("/api/hospital/profile", methods=["POST"])
+def update_hospital_profile():
+    err = _hospital_required()
+    if err:
+        return err
+    data    = request.get_json(force=True) or {}
+    allowed = {"hospital_name", "location", "research_conditions"}
+    kwargs  = {k: v for k, v in data.items() if k in allowed}
+    db.update_hospital_profile(session["hospital_id"], **kwargs)
+    return jsonify(_pub_hospital(db.get_hospital_by_id(session["hospital_id"])))
+
+
+@app.route("/api/hospital/trials", methods=["GET"])
+def get_hospital_trials():
+    err = _hospital_required()
+    if err:
+        return err
+    if not _ready():
+        return jsonify({"trials": [], "message": "System still loading"}), 202
+    hospital = db.get_hospital_by_id(session["hospital_id"])
+    if not hospital:
+        return jsonify({"trials": []}), 404
+    trials = pipeline.trials_for_hospital(
+        hospital.get("hospital_name", ""),
+        hospital.get("location", ""),
+        hospital.get("research_conditions", []),
+        top_k=20,
+    )
+    return jsonify({"trials": trials, "total": len(trials)})
+
+
+@app.route("/api/hospital/patients", methods=["GET"])
+def get_hospital_patients():
+    err = _hospital_required()
+    if err:
+        return err
+    condition         = request.args.get("condition", "").strip()
+    include_connected = request.args.get("include_connected", "0") == "1"
+    patients          = db.get_open_patients_for_hospital(
+        session["hospital_id"], condition, include_connected=include_connected
+    )
+    return jsonify({"patients": patients})
+
+
+@app.route("/api/hospital/connect", methods=["POST"])
+def hospital_connect():
+    err = _hospital_required()
+    if err:
+        return err
+    data        = request.get_json(force=True) or {}
+    patient_id  = data.get("patient_id", "")
+    trial_id    = data.get("trial_id", "")
+    trial_title = data.get("trial_title", "")
+    message     = data.get("message", "")
+    if not patient_id:
+        return jsonify({"error": "patient_id required"}), 400
+    conn = db.create_connection(
+        patient_id, session["hospital_id"], trial_id, trial_title,
+        initiated_by="hospital", message=message,
+    )
+    if conn is None:
+        return jsonify({"error": "A connection with this patient already exists"}), 409
+    return jsonify({"success": True, "connection": conn})
+
+
+@app.route("/api/hospital/connections", methods=["GET"])
+def get_hospital_connections():
+    err = _hospital_required()
+    if err:
+        return err
+    return jsonify({"connections": db.get_hospital_connections(session["hospital_id"])})
+
+
+@app.route("/api/hospital/connections/<cid>/status", methods=["PUT"])
+def update_hospital_connection_status(cid):
+    err = _hospital_required()
+    if err:
+        return err
+    data   = request.get_json(force=True) or {}
+    status = data.get("status", "")
+    if status not in ("pending", "accepted", "rejected", "completed"):
+        return jsonify({"error": "Invalid status"}), 400
+    db.update_connection_status(cid, status)
+    return jsonify({"success": True})
+
+
+# ---------------------------------------------------------------------------
+# Connection Messages
+# ---------------------------------------------------------------------------
+
+@app.route("/api/patient/connections/<cid>/messages", methods=["GET"])
+def get_patient_connection_messages(cid):
+    err = _patient_required()
+    if err:
+        return err
+    conn = db.get_connection(cid)
+    if not conn or conn["patient_id"] != session["patient_id"]:
+        return jsonify({"error": "Not found"}), 404
+    db.mark_messages_read(cid, "patient")
+    return jsonify({"messages": db.get_connection_messages(cid)})
+
+
+@app.route("/api/patient/connections/<cid>/messages", methods=["POST"])
+def post_patient_connection_message(cid):
+    err = _patient_required()
+    if err:
+        return err
+    conn = db.get_connection(cid)
+    if not conn or conn["patient_id"] != session["patient_id"]:
+        return jsonify({"error": "Not found"}), 404
+    body = (request.get_json(force=True) or {}).get("body", "").strip()
+    if not body:
+        return jsonify({"error": "Message body required"}), 400
+    msg = db.create_connection_message(cid, "patient", session["patient_id"], body)
+    return jsonify({"success": True, "message": msg})
+
+
+@app.route("/api/hospital/connections/<cid>/messages", methods=["GET"])
+def get_hospital_connection_messages(cid):
+    err = _hospital_required()
+    if err:
+        return err
+    conn = db.get_connection(cid)
+    if not conn or conn["hospital_id"] != session["hospital_id"]:
+        return jsonify({"error": "Not found"}), 404
+    db.mark_messages_read(cid, "hospital")
+    return jsonify({"messages": db.get_connection_messages(cid)})
+
+
+@app.route("/api/hospital/connections/<cid>/messages", methods=["POST"])
+def post_hospital_connection_message(cid):
+    err = _hospital_required()
+    if err:
+        return err
+    conn = db.get_connection(cid)
+    if not conn or conn["hospital_id"] != session["hospital_id"]:
+        return jsonify({"error": "Not found"}), 404
+    body = (request.get_json(force=True) or {}).get("body", "").strip()
+    if not body:
+        return jsonify({"error": "Message body required"}), 400
+    msg = db.create_connection_message(cid, "hospital", session["hospital_id"], body)
+    return jsonify({"success": True, "message": msg})
+
+
+# ---------------------------------------------------------------------------
+# Patient Documents
+# ---------------------------------------------------------------------------
+
+@app.route("/api/patient/documents", methods=["GET"])
+def get_patient_documents():
+    err = _patient_required()
+    if err:
+        return err
+    patient = db.get_patient_by_id(session["patient_id"])
+    return jsonify({"documents": patient.get("documents", []) or []})
+
+
+@app.route("/api/patient/documents", methods=["POST"])
+def upload_patient_document():
+    err = _patient_required()
+    if err:
+        return err
+    if "file" not in request.files or not request.files["file"].filename:
+        return jsonify({"error": "No file selected"}), 400
+    f   = request.files["file"]
+    ext = Path(f.filename).suffix.lower()
+    if ext not in ALLOWED_EXTENSIONS:
+        return jsonify({"error": f"Allowed types: {', '.join(sorted(ALLOWED_EXTENSIONS))}"}), 400
+
+    pid     = session["patient_id"]
+    doc_dir = UPLOAD_DIR / pid
+    doc_dir.mkdir(parents=True, exist_ok=True)
+
+    doc_id    = str(uuid.uuid4())
+    safe_name = secure_filename(f.filename)
+    save_path = doc_dir / f"{doc_id}_{safe_name}"
+    f.save(str(save_path))
+
+    doc_meta = {
+        "id":          doc_id,
+        "filename":    safe_name,
+        "path":        str(save_path.relative_to(Path(__file__).parent)),
+        "uploaded_at": datetime.now().isoformat(),
+        "mime_type":   mimetypes.guess_type(f.filename)[0] or "application/octet-stream",
+        "size_bytes":  save_path.stat().st_size,
+    }
+    patient = db.get_patient_by_id(pid)
+    docs    = list(patient.get("documents", []) or [])
+    docs.append(doc_meta)
+    db.update_patient_profile(pid, documents=docs)
+    return jsonify({"success": True, "document": doc_meta})
+
+
+@app.route("/api/patient/documents/<doc_id>", methods=["DELETE"])
+def delete_patient_document(doc_id):
+    err = _patient_required()
+    if err:
+        return err
+    pid     = session["patient_id"]
+    patient = db.get_patient_by_id(pid)
+    docs    = list(patient.get("documents", []) or [])
+    target  = next((d for d in docs if d["id"] == doc_id), None)
+    if not target:
+        return jsonify({"error": "Not found"}), 404
+    try:
+        fpath = Path(__file__).parent / target["path"]
+        if fpath.exists():
+            fpath.unlink()
+    except Exception:
+        pass
+    db.update_patient_profile(pid, documents=[d for d in docs if d["id"] != doc_id])
+    return jsonify({"success": True})
+
+
+@app.route("/api/patient/documents/<doc_id>/download")
+def download_patient_document(doc_id):
+    err = _patient_required()
+    if err:
+        return err
+    patient = db.get_patient_by_id(session["patient_id"])
+    docs    = list(patient.get("documents", []) or [])
+    target  = next((d for d in docs if d["id"] == doc_id), None)
+    if not target:
+        return jsonify({"error": "Not found"}), 404
+    fpath = Path(__file__).parent / target["path"]
+    if not fpath.exists():
+        return jsonify({"error": "File not found on disk"}), 404
+    return send_file(str(fpath), download_name=target["filename"], as_attachment=True)
+
+
+# ---------------------------------------------------------------------------
+# Inbox
+# ---------------------------------------------------------------------------
+
+@app.route("/api/hospital/inbox")
+def get_hospital_inbox():
+    err = _hospital_required()
+    if err:
+        return err
+    return jsonify({"threads": db.get_hospital_inbox_threads(session["hospital_id"])})
+
+
+@app.route("/api/patient/inbox")
+def get_patient_inbox():
+    err = _patient_required()
+    if err:
+        return err
+    return jsonify({"threads": db.get_patient_inbox_threads(session["patient_id"])})
+
+
+# ---------------------------------------------------------------------------
+# Shared pipeline API
+# ---------------------------------------------------------------------------
+
+@app.route("/api/status")
+def api_status():
+    if _boot_error:
+        return jsonify({"ready": False, "error": _boot_error}), 500
+    if not _ready():
+        return jsonify({"ready": False,
+                        "message": "Loading data and training model…"}), 202
+    return jsonify({"ready": True, "stats": pipeline.stats})
+
+
+@app.route("/api/conditions/autocomplete")
+def api_conditions_autocomplete():
+    if not _ready():
+        return jsonify({"results": []})
+    q = request.args.get("q", "").strip()
+    if len(q) < 2:
+        return jsonify({"results": []})
+    return jsonify({"results": pipeline.condition_autocomplete(q, limit=15)})
+
+
+# ---------------------------------------------------------------------------
+# Main
+# ---------------------------------------------------------------------------
+if __name__ == "__main__":
+    port = int(os.environ.get("PORT", 5000))
+    print("=" * 60)
+    print(" Second Life — Clinical Trial Matching")
+    print(f" http://localhost:{port}")
+    print("=" * 60)
+    app.run(host="0.0.0.0", debug=False, port=port, use_reloader=False)

claude_handoff.md

@@ -0,0 +1,65 @@
+# Claude Handoff
+
+## Current App Shape
+
+Multi-page Flask app. All files committed + local session changes applied.
+
+Active runtime files:
+
+- `app.py`: Flask routes and portal/API logic
+- `pipeline.py`: data loading, feature engineering, model training, trial matching
+- `database.py`: SQLite accounts, interests, connections
+- `templates/landing.html`: login / registration
+- `templates/patient.html`: patient portal
+- `templates/hospital.html`: hospital portal
+
+`templates/index.html` deleted.
+
+## Status: All Fixes Complete
+
+The following work is done and local (not yet committed to Git):
+
+### From Previous Session (before rate limit)
+
+1. Hospital suggestion tiered logic (Tier 1 facility-name Jaccard, Tier 2 same-state, Tier 3 condition overlap, Tier 4 fallback)
+2. Duplicate connection prevention (schema UNIQUE + COALESCE index)
+3. Hospital patient feed excludes already-connected patients
+4. Demo patient seeded with `open_to_trials = 1`
+5. Hospital profile editing tab (name, location, research conditions)
+6. Patient-facing model disclaimer on trial results
+7. Trial site info (lead site + location) surfaced in patient UI
+8. Hospital registration collects `research_conditions`
+
+### From Current Session
+
+9. **Patient connect modal**: Tier 1 hospitals shown under "Verified Trial Sites" (green header); Tiers 2/3/4 shown under "Related Hospitals — not confirmed trial sites" (grey header). Two visually separated sections.
+10. **hospital.html bug fix**: `btn-close-white` → `btn-close` on `bg-info` profile condition tags (white X was invisible on light-blue background).
+11. **landing.html**: Enter key now submits login/register forms (all username + password inputs).
+12. **patient.html**: System status banner auto-clears — polls `/api/status` every 5s until pipeline is ready, then stops.
+
+## Decision Log
+
+**Hospital matching mode: BROAD**
+- Tier 1 = verified site (Jaccard name match ≥ 0.25)
+- Tiers 2+3 = related hospitals (same state / condition overlap)
+- Visual separation in modal so user sees which is which
+- Reason: safer for demo — modal won't be empty; stronger matches still shown first
+
+## Known Limitation (not a bug, by design)
+
+Model trains on synthetic/rule-based labels. Disclaimer shown in UI. Real fix requires clinician-reviewed labels or historical screening decisions — out of scope for DSCI 5260.
+
+## Demo Credentials
+
+| Role | Username | Password |
+|------|----------|----------|
+| Patient | john_doe | pass123 |
+| Hospital | mgh | mgh123 |
+
+## Recommended Test Flow
+
+See "What to Test" section returned at end of last Claude session.
+
+## Git Note
+
+Last GitHub push: `ff67a12` — all local work above is uncommitted.

database.py

@@ -0,0 +1,702 @@
+"""
+Second Life — SQLite database layer
+Handles patient accounts, hospital accounts, trial interests, and connections.
+"""
+
+import csv
+import hashlib
+import json
+import sqlite3
+import uuid
+from datetime import datetime
+from pathlib import Path
+
+BASE_DIR = Path(__file__).parent
+DB_PATH = BASE_DIR / "secondlife.db"
+PATIENT_DETAILS_PATH = BASE_DIR / "Final Patients Synthea Data" / "patients_details.csv"
+PATIENT_CONDITIONS_PATH = BASE_DIR / "Final Patients Synthea Data" / "final_patients_conditions.csv"
+PATIENT_MEDICATIONS_PATH = BASE_DIR / "Final Patients Synthea Data" / "patients_medications.csv"
+DATASET_PATIENT_SEED_COUNT = 20
+
+# ---------------------------------------------------------------------------
+# Schema
+# ---------------------------------------------------------------------------
+_SCHEMA = """
+CREATE TABLE IF NOT EXISTS patient_accounts (
+    id              TEXT PRIMARY KEY,
+    username        TEXT UNIQUE NOT NULL,
+    password_hash   TEXT NOT NULL,
+    synthea_id      TEXT,
+    first_name      TEXT,
+    last_name       TEXT,
+    dob             TEXT,
+    gender          TEXT,
+    address         TEXT,
+    conditions      TEXT DEFAULT '[]',
+    medications     TEXT DEFAULT '[]',
+    documents       TEXT DEFAULT '[]',
+    open_to_trials  INTEGER DEFAULT 0,
+    created_at      TEXT DEFAULT CURRENT_TIMESTAMP
+);
+
+CREATE TABLE IF NOT EXISTS hospital_accounts (
+    id                  TEXT PRIMARY KEY,
+    username            TEXT UNIQUE NOT NULL,
+    password_hash       TEXT NOT NULL,
+    hospital_name       TEXT NOT NULL,
+    location            TEXT,
+    research_conditions TEXT DEFAULT '[]',
+    created_at          TEXT DEFAULT CURRENT_TIMESTAMP
+);
+
+CREATE TABLE IF NOT EXISTS patient_trial_interests (
+    id          TEXT PRIMARY KEY,
+    patient_id  TEXT NOT NULL,
+    trial_id    TEXT NOT NULL,
+    trial_title TEXT,
+    match_score REAL,
+    status      TEXT DEFAULT 'interested',
+    created_at  TEXT DEFAULT CURRENT_TIMESTAMP,
+    UNIQUE(patient_id, trial_id)
+);
+
+CREATE TABLE IF NOT EXISTS connections (
+    id              TEXT PRIMARY KEY,
+    patient_id      TEXT NOT NULL,
+    hospital_id     TEXT NOT NULL,
+    trial_id        TEXT,
+    trial_title     TEXT,
+    initiated_by    TEXT DEFAULT 'patient',
+    status          TEXT DEFAULT 'pending',
+    message         TEXT,
+    created_at      TEXT DEFAULT CURRENT_TIMESTAMP,
+    UNIQUE(patient_id, hospital_id, trial_id)
+);
+
+CREATE TABLE IF NOT EXISTS connection_messages (
+    id              TEXT PRIMARY KEY,
+    connection_id   TEXT NOT NULL,
+    sender_role     TEXT NOT NULL,
+    sender_id       TEXT NOT NULL,
+    body            TEXT NOT NULL,
+    created_at      TEXT DEFAULT CURRENT_TIMESTAMP,
+    is_read         INTEGER DEFAULT 0
+);
+"""
+
+# Run after schema to add constraint to pre-existing DBs that lack it
+_POST_SCHEMA = """
+CREATE UNIQUE INDEX IF NOT EXISTS idx_conn_unique
+    ON connections(patient_id, hospital_id, COALESCE(trial_id, ''));
+CREATE INDEX IF NOT EXISTS idx_msgs_conn
+    ON connection_messages(connection_id, created_at);
+"""
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+def _conn():
+    c = sqlite3.connect(DB_PATH)
+    c.row_factory = sqlite3.Row
+    return c
+
+
+def _hash(password: str) -> str:
+    return hashlib.sha256(password.encode()).hexdigest()
+
+
+def _row_to_dict(row) -> dict:
+    if row is None:
+        return None
+    d = dict(row)
+    for k in ("conditions", "medications", "documents", "research_conditions"):
+        if k in d and isinstance(d[k], str):
+            try:
+                d[k] = json.loads(d[k])
+            except Exception:
+                d[k] = []
+    return d
+
+
+# ---------------------------------------------------------------------------
+# Init
+# ---------------------------------------------------------------------------
+
+def init_db():
+    with _conn() as c:
+        c.executescript(_SCHEMA)
+        try:
+            c.executescript(_POST_SCHEMA)
+        except Exception:
+            pass  # index may already exist with different definition
+    _seed_demo_data()
+
+
+def _seed_demo_data():
+    """Seed demo hospital and patient accounts if not already present."""
+    with _conn() as c:
+        # Demo hospitals
+        hospitals = [
+            ("mgh", "mgh123", "Massachusetts General Hospital", "Boston, MA",
+             ["hypertension", "diabetes", "cardiac arrest", "stroke"]),
+            ("cleveland", "clinic123", "Cleveland Clinic", "Cleveland, OH",
+             ["coronary artery disease", "myocardial infarction", "heart failure"]),
+            ("jhopkins", "johns123", "Johns Hopkins Hospital", "Baltimore, MD",
+             ["cancer", "non-small cell lung cancer", "malignant tumor of colon"]),
+        ]
+        for uname, pwd, name, loc, conds in hospitals:
+            existing = c.execute(
+                "SELECT id FROM hospital_accounts WHERE username=?", (uname,)
+            ).fetchone()
+            if not existing:
+                c.execute(
+                    "INSERT INTO hospital_accounts VALUES (?,?,?,?,?,?,?)",
+                    (str(uuid.uuid4()), uname, _hash(pwd), name, loc,
+                     json.dumps(conds), datetime.now().isoformat())
+                )
+
+        # Demo patient accounts
+        patients = [
+            ("john_doe",   "pass123", None, "John",   "Doe",    "1965-03-12", "M",
+             "123 Main St Boston MA 02101 US",
+             ["hypertension", "diabetes", "myocardial infarction"],
+             ["metformin", "lisinopril"], 1),
+            ("jane_smith", "pass123", None, "Jane",   "Smith",  "1978-07-22", "F",
+             "456 Oak Ave Cambridge MA 02139 US",
+             ["asthma", "atopic dermatitis", "seasonal allergic rhinitis"],
+             ["albuterol", "fluticasone"], 1),
+            ("bob_jones",  "pass123", None, "Robert", "Jones",  "1955-11-05", "M",
+             "789 Pine Rd Cleveland OH 44106 US",
+             ["coronary artery disease", "hypertension", "chronic pain"],
+             ["atorvastatin", "aspirin"], 1),
+            ("alice_brown","pass123", None, "Alice",  "Brown",  "1972-06-14", "F",
+             "101 Elm St Baltimore MD 21201 US",
+             ["non-small cell lung cancer", "stroke"],
+             ["erlotinib"], 1),
+            ("david_chen", "pass123", None, "David",  "Chen",   "1948-09-30", "M",
+             "202 Oak Blvd Chicago IL 60601 US",
+             ["diabetes", "osteoporosis", "coronary artery disease"],
+             ["insulin", "alendronate"], 1),
+        ]
+        for uname, pwd, syn_id, fn, ln, dob, gend, addr, conds, meds, open_trials in patients:
+            existing = c.execute(
+                "SELECT id FROM patient_accounts WHERE username=?", (uname,)
+            ).fetchone()
+            if not existing:
+                c.execute(
+                    """INSERT INTO patient_accounts
+                       (id,username,password_hash,synthea_id,first_name,last_name,
+                        dob,gender,address,conditions,medications,open_to_trials,created_at)
+                       VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?)""",
+                    (str(uuid.uuid4()), uname, _hash(pwd), syn_id, fn, ln,
+                     dob, gend, addr,
+                     json.dumps(conds), json.dumps(meds), open_trials,
+                     datetime.now().isoformat())
+                )
+            else:
+                # Ensure existing demo patients are open_to_trials=1
+                c.execute(
+                    "UPDATE patient_accounts SET open_to_trials=1 WHERE username=?", (uname,)
+                )
+        _seed_dataset_patients(c, max_patients=DATASET_PATIENT_SEED_COUNT)
+        c.commit()
+
+
+def _seed_dataset_patients(c, max_patients: int = 20):
+    """
+    Seed a small set of real Synthea patients into the portal so the hospital
+    dashboards are not limited to hand-made demo accounts.
+    """
+    if not (PATIENT_DETAILS_PATH.exists() and PATIENT_CONDITIONS_PATH.exists()
+            and PATIENT_MEDICATIONS_PATH.exists()):
+        return
+
+    existing_dataset_count = c.execute(
+        "SELECT COUNT(*) FROM patient_accounts WHERE synthea_id IS NOT NULL"
+    ).fetchone()[0]
+    if existing_dataset_count >= max_patients:
+        return
+
+    existing_synthea_ids = {
+        row[0] for row in c.execute(
+            "SELECT synthea_id FROM patient_accounts WHERE synthea_id IS NOT NULL"
+        ).fetchall()
+    }
+    needed = max_patients - existing_dataset_count
+    selected = {}
+
+    with PATIENT_DETAILS_PATH.open("r", encoding="utf-8", newline="") as fh:
+        reader = csv.DictReader(fh)
+        for row in reader:
+            if len(selected) >= needed:
+                break
+            sid = (row.get("Patient_ID") or "").strip()
+            if not sid or sid in existing_synthea_ids:
+                continue
+            if (row.get("Death_Date") or "").strip():
+                continue
+
+            first_name = (row.get("First_Name") or "").strip() or "Patient"
+            last_name = (row.get("Last_Name") or "").strip() or sid[:6]
+            dob = _normalize_dataset_date(row.get("Birth_Date", ""))
+            gender = (row.get("Gender") or "").strip()
+            address = (row.get("Address") or "").strip()
+            username = f"synthea_{sid[:8].lower()}"
+            selected[sid] = {
+                "username": username,
+                "first_name": first_name,
+                "last_name": last_name,
+                "dob": dob,
+                "gender": gender,
+                "address": address,
+                "conditions": [],
+                "medications": [],
+            }
+
+    if not selected:
+        return
+
+    with PATIENT_CONDITIONS_PATH.open("r", encoding="utf-8", newline="") as fh:
+        reader = csv.DictReader(fh)
+        for row in reader:
+            sid = (row.get("Patient_ID") or "").strip()
+            if sid not in selected:
+                continue
+            cond = (row.get("Condition_Name") or "").strip().lower()
+            if cond and cond not in selected[sid]["conditions"]:
+                selected[sid]["conditions"].append(cond)
+                if len(selected[sid]["conditions"]) >= 8:
+                    continue
+
+    with PATIENT_MEDICATIONS_PATH.open("r", encoding="utf-8", newline="") as fh:
+        reader = csv.DictReader(fh)
+        for row in reader:
+            sid = (row.get("Patient_ID") or "").strip()
+            if sid not in selected:
+                continue
+            med = (row.get("Medication_Name") or "").strip()
+            if med and med not in selected[sid]["medications"]:
+                selected[sid]["medications"].append(med)
+                if len(selected[sid]["medications"]) >= 6:
+                    continue
+
+    for sid, patient in selected.items():
+        c.execute(
+            """INSERT OR IGNORE INTO patient_accounts
+               (id,username,password_hash,synthea_id,first_name,last_name,
+                dob,gender,address,conditions,medications,open_to_trials,created_at)
+               VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?)""",
+            (
+                str(uuid.uuid4()),
+                patient["username"],
+                _hash("pass123"),
+                sid,
+                patient["first_name"],
+                patient["last_name"],
+                patient["dob"],
+                patient["gender"],
+                patient["address"],
+                json.dumps(patient["conditions"]),
+                json.dumps(patient["medications"]),
+                1,
+                datetime.now().isoformat(),
+            ),
+        )
+
+
+def _normalize_dataset_date(raw: str) -> str:
+    raw = (raw or "").strip()
+    if not raw:
+        return ""
+    try:
+        return datetime.strptime(raw, "%d-%m-%Y").strftime("%Y-%m-%d")
+    except ValueError:
+        return raw
+
+
+# ---------------------------------------------------------------------------
+# Patient CRUD
+# ---------------------------------------------------------------------------
+
+def create_patient(username, password, first_name="", last_name="",
+                   dob="", gender="", address="", synthea_id=None) -> dict | None:
+    pid = str(uuid.uuid4())
+    try:
+        with _conn() as c:
+            c.execute(
+                """INSERT INTO patient_accounts
+                   (id,username,password_hash,synthea_id,first_name,last_name,
+                    dob,gender,address,created_at)
+                   VALUES (?,?,?,?,?,?,?,?,?,?)""",
+                (pid, username, _hash(password), synthea_id,
+                 first_name, last_name, dob, gender, address,
+                 datetime.now().isoformat())
+            )
+            c.commit()
+        return get_patient_by_id(pid)
+    except sqlite3.IntegrityError:
+        return None   # username taken
+
+
+def get_patient_by_id(pid: str) -> dict | None:
+    with _conn() as c:
+        return _row_to_dict(c.execute(
+            "SELECT * FROM patient_accounts WHERE id=?", (pid,)
+        ).fetchone())
+
+
+def get_patient_by_username(username: str) -> dict | None:
+    with _conn() as c:
+        return _row_to_dict(c.execute(
+            "SELECT * FROM patient_accounts WHERE username=?", (username,)
+        ).fetchone())
+
+
+def authenticate_patient(username: str, password: str) -> dict | None:
+    p = get_patient_by_username(username)
+    if p and p["password_hash"] == _hash(password):
+        return p
+    return None
+
+
+def update_patient_profile(pid: str, **kwargs):
+    allowed = {"first_name", "last_name", "dob", "gender", "address",
+               "conditions", "medications", "documents", "open_to_trials", "synthea_id"}
+    fields, vals = [], []
+    for k, v in kwargs.items():
+        if k in allowed:
+            fields.append(f"{k}=?")
+            vals.append(json.dumps(v) if isinstance(v, (list, dict)) else v)
+    if not fields:
+        return
+    vals.append(pid)
+    with _conn() as c:
+        c.execute(f"UPDATE patient_accounts SET {', '.join(fields)} WHERE id=?", vals)
+        c.commit()
+
+
+# ---------------------------------------------------------------------------
+# Hospital CRUD
+# ---------------------------------------------------------------------------
+
+def create_hospital(username, password, hospital_name, location="",
+                    research_conditions=None) -> dict | None:
+    hid = str(uuid.uuid4())
+    try:
+        with _conn() as c:
+            c.execute(
+                """INSERT INTO hospital_accounts
+                   (id,username,password_hash,hospital_name,location,research_conditions,created_at)
+                   VALUES (?,?,?,?,?,?,?)""",
+                (hid, username, _hash(password), hospital_name, location,
+                 json.dumps(research_conditions or []),
+                 datetime.now().isoformat())
+            )
+            c.commit()
+        return get_hospital_by_id(hid)
+    except sqlite3.IntegrityError:
+        return None
+
+
+def get_hospital_by_id(hid: str) -> dict | None:
+    with _conn() as c:
+        return _row_to_dict(c.execute(
+            "SELECT * FROM hospital_accounts WHERE id=?", (hid,)
+        ).fetchone())
+
+
+def get_hospital_by_username(username: str) -> dict | None:
+    with _conn() as c:
+        return _row_to_dict(c.execute(
+            "SELECT * FROM hospital_accounts WHERE username=?", (username,)
+        ).fetchone())
+
+
+def authenticate_hospital(username: str, password: str) -> dict | None:
+    h = get_hospital_by_username(username)
+    if h and h["password_hash"] == _hash(password):
+        return h
+    return None
+
+
+def get_all_hospitals() -> list:
+    with _conn() as c:
+        rows = c.execute(
+            "SELECT id, hospital_name, location, research_conditions FROM hospital_accounts"
+        ).fetchall()
+    return [_row_to_dict(r) for r in rows]
+
+
+def update_hospital_profile(hid: str, **kwargs):
+    allowed = {"hospital_name", "location", "research_conditions"}
+    fields, vals = [], []
+    for k, v in kwargs.items():
+        if k in allowed:
+            fields.append(f"{k}=?")
+            vals.append(json.dumps(v) if isinstance(v, (list, dict)) else v)
+    if not fields:
+        return
+    vals.append(hid)
+    with _conn() as c:
+        c.execute(f"UPDATE hospital_accounts SET {', '.join(fields)} WHERE id=?", vals)
+        c.commit()
+
+
+# ---------------------------------------------------------------------------
+# Trial Interests
+# ---------------------------------------------------------------------------
+
+def save_trial_interest(patient_id, trial_id, trial_title, match_score):
+    iid = str(uuid.uuid4())
+    with _conn() as c:
+        c.execute(
+            """INSERT OR REPLACE INTO patient_trial_interests
+               (id,patient_id,trial_id,trial_title,match_score,status,created_at)
+               VALUES (?,?,?,?,?,'interested',?)""",
+            (iid, patient_id, trial_id, trial_title[:200],
+             match_score, datetime.now().isoformat())
+        )
+        c.commit()
+
+
+def get_patient_interests(patient_id: str) -> list:
+    with _conn() as c:
+        rows = c.execute(
+            """SELECT * FROM patient_trial_interests
+               WHERE patient_id=? ORDER BY match_score DESC""",
+            (patient_id,)
+        ).fetchall()
+    return [dict(r) for r in rows]
+
+
+def withdraw_interest(patient_id, trial_id):
+    with _conn() as c:
+        c.execute(
+            """UPDATE patient_trial_interests SET status='withdrawn'
+               WHERE patient_id=? AND trial_id=?""",
+            (patient_id, trial_id)
+        )
+        c.commit()
+
+
+# ---------------------------------------------------------------------------
+# Connections
+# ---------------------------------------------------------------------------
+
+def connection_exists(patient_id: str, hospital_id: str, trial_id: str) -> bool:
+    with _conn() as c:
+        row = c.execute(
+            """SELECT id FROM connections
+               WHERE patient_id=? AND hospital_id=? AND COALESCE(trial_id,'')=COALESCE(?,'')""",
+            (patient_id, hospital_id, trial_id)
+        ).fetchone()
+    return row is not None
+
+
+def create_connection(patient_id, hospital_id, trial_id, trial_title,
+                      initiated_by="patient", message="") -> dict | None:
+    """Returns None if a connection for this (patient, hospital, trial) already exists."""
+    if connection_exists(patient_id, hospital_id, trial_id):
+        return None
+    cid = str(uuid.uuid4())
+    try:
+        with _conn() as c:
+            c.execute(
+                """INSERT INTO connections
+                   (id,patient_id,hospital_id,trial_id,trial_title,
+                    initiated_by,status,message,created_at)
+                   VALUES (?,?,?,?,?,?,  'pending',?,?)""",
+                (cid, patient_id, hospital_id, trial_id, trial_title[:200] if trial_title else "",
+                 initiated_by, message, datetime.now().isoformat())
+            )
+            c.commit()
+    except sqlite3.IntegrityError:
+        return None  # race condition — already inserted
+    return get_connection(cid)
+
+
+def get_connection(cid: str) -> dict | None:
+    with _conn() as c:
+        return _row_to_dict(c.execute(
+            "SELECT * FROM connections WHERE id=?", (cid,)
+        ).fetchone())
+
+
+def get_patient_connections(patient_id: str) -> list:
+    with _conn() as c:
+        rows = c.execute(
+            """SELECT c.*, h.hospital_name, h.location as hospital_location
+               FROM connections c
+               JOIN hospital_accounts h ON c.hospital_id=h.id
+               WHERE c.patient_id=? ORDER BY c.created_at DESC""",
+            (patient_id,)
+        ).fetchall()
+    return [dict(r) for r in rows]
+
+
+def get_hospital_connections(hospital_id: str) -> list:
+    with _conn() as c:
+        rows = c.execute(
+            """SELECT c.*,
+                      p.first_name, p.last_name, p.gender, p.dob,
+                      p.conditions, p.address
+               FROM connections c
+               JOIN patient_accounts p ON c.patient_id=p.id
+               WHERE c.hospital_id=? ORDER BY c.created_at DESC""",
+            (hospital_id,)
+        ).fetchall()
+    result = []
+    for r in rows:
+        d = dict(r)
+        for k in ("conditions",):
+            if isinstance(d.get(k), str):
+                try:
+                    d[k] = json.loads(d[k])
+                except Exception:
+                    d[k] = []
+        result.append(d)
+    return result
+
+
+def update_connection_status(cid: str, status: str):
+    with _conn() as c:
+        c.execute("UPDATE connections SET status=? WHERE id=?", (status, cid))
+        c.commit()
+
+
+def get_open_patients_for_hospital(hospital_id: str, condition_filter: str = "",
+                                    include_connected: bool = False) -> list:
+    """
+    Returns patients who are open_to_trials=1, optionally filtered by condition.
+    When include_connected=False (default), excludes patients already connected to this hospital.
+    """
+    with _conn() as c:
+        if include_connected:
+            rows = c.execute(
+                """SELECT id, first_name, last_name, gender, dob, address, conditions
+                   FROM patient_accounts WHERE open_to_trials=1"""
+            ).fetchall()
+        else:
+            rows = c.execute(
+                """SELECT id, first_name, last_name, gender, dob, address, conditions
+                   FROM patient_accounts
+                   WHERE open_to_trials=1
+                     AND id NOT IN (
+                         SELECT DISTINCT patient_id FROM connections
+                         WHERE hospital_id=?
+                     )""",
+                (hospital_id,)
+            ).fetchall()
+
+    result = []
+    cf = condition_filter.lower().strip()
+    for r in rows:
+        d = dict(r)
+        try:
+            d["conditions"] = json.loads(d["conditions"]) if isinstance(d["conditions"], str) else d["conditions"]
+        except Exception:
+            d["conditions"] = []
+        if cf and not any(cf in cond.lower() for cond in d["conditions"]):
+            continue
+        result.append(d)
+    return result
+
+
+# ---------------------------------------------------------------------------
+# Connection Messages
+# ---------------------------------------------------------------------------
+
+def get_connection_messages(connection_id: str) -> list:
+    with _conn() as c:
+        rows = c.execute(
+            """SELECT * FROM connection_messages
+               WHERE connection_id=? ORDER BY created_at ASC""",
+            (connection_id,)
+        ).fetchall()
+    return [dict(r) for r in rows]
+
+
+def create_connection_message(connection_id: str, sender_role: str,
+                               sender_id: str, body: str) -> dict:
+    mid = str(uuid.uuid4())
+    with _conn() as c:
+        c.execute(
+            """INSERT INTO connection_messages
+               (id, connection_id, sender_role, sender_id, body, created_at)
+               VALUES (?, ?, ?, ?, ?, ?)""",
+            (mid, connection_id, sender_role, sender_id,
+             body[:2000], datetime.now().isoformat())
+        )
+        c.commit()
+    with _conn() as c:
+        row = c.execute(
+            "SELECT * FROM connection_messages WHERE id=?", (mid,)
+        ).fetchone()
+    return dict(row) if row else {}
+
+
+def mark_messages_read(connection_id: str, reader_role: str):
+    with _conn() as c:
+        c.execute(
+            """UPDATE connection_messages SET is_read=1
+               WHERE connection_id=? AND sender_role != ?""",
+            (connection_id, reader_role)
+        )
+        c.commit()
+
+
+def unread_count(connection_id: str, reader_role: str) -> int:
+    with _conn() as c:
+        row = c.execute(
+            """SELECT COUNT(*) AS cnt FROM connection_messages
+               WHERE connection_id=? AND sender_role != ? AND is_read=0""",
+            (connection_id, reader_role)
+        ).fetchone()
+    return row["cnt"] if row else 0
+
+
+def get_hospital_inbox_threads(hospital_id: str) -> list:
+    """All connection threads for a hospital, ordered by most recent activity."""
+    with _conn() as c:
+        rows = c.execute(
+            """SELECT
+                   c.id, c.patient_id, c.trial_title, c.status, c.created_at,
+                   p.first_name, p.last_name,
+                   (SELECT body FROM connection_messages
+                    WHERE connection_id=c.id ORDER BY created_at DESC LIMIT 1) AS last_message,
+                   (SELECT created_at FROM connection_messages
+                    WHERE connection_id=c.id ORDER BY created_at DESC LIMIT 1) AS last_message_at,
+                   (SELECT COUNT(*) FROM connection_messages
+                    WHERE connection_id=c.id AND sender_role='patient' AND is_read=0) AS unread_count
+               FROM connections c
+               JOIN patient_accounts p ON c.patient_id=p.id
+               WHERE c.hospital_id=?""",
+            (hospital_id,)
+        ).fetchall()
+    threads = [dict(r) for r in rows]
+    threads.sort(key=lambda t: t.get("last_message_at") or t["created_at"], reverse=True)
+    return threads
+
+
+def get_patient_inbox_threads(patient_id: str) -> list:
+    """All connection threads for a patient, ordered by most recent activity."""
+    with _conn() as c:
+        rows = c.execute(
+            """SELECT
+                   c.id, c.hospital_id, c.trial_title, c.status, c.created_at,
+                   h.hospital_name,
+                   (SELECT body FROM connection_messages
+                    WHERE connection_id=c.id ORDER BY created_at DESC LIMIT 1) AS last_message,
+                   (SELECT created_at FROM connection_messages
+                    WHERE connection_id=c.id ORDER BY created_at DESC LIMIT 1) AS last_message_at,
+                   (SELECT COUNT(*) FROM connection_messages
+                    WHERE connection_id=c.id AND sender_role='hospital' AND is_read=0) AS unread_count
+               FROM connections c
+               JOIN hospital_accounts h ON c.hospital_id=h.id
+               WHERE c.patient_id=?""",
+            (patient_id,)
+        ).fetchall()
+    threads = [dict(r) for r in rows]
+    threads.sort(key=lambda t: t.get("last_message_at") or t["created_at"], reverse=True)
+    return threads

hf_space_bootstrap.py

@@ -0,0 +1,58 @@
+"""Hugging Face Space startup helper.
+
+The Flask app expects two local dataset folders. In the Space, those folders are
+downloaded from a separate Hugging Face Dataset repo before importing app.py,
+because app.py starts the pipeline boot thread at import time.
+"""
+
+from __future__ import annotations
+
+import os
+import zipfile
+from pathlib import Path
+
+from huggingface_hub import hf_hub_download
+
+
+BASE_DIR = Path(__file__).resolve().parent
+DATASET_REPO = os.environ.get("HF_DATASET_REPO", "MrNoOne07/second-life-data")
+
+DATASET_ZIPS = [
+    ("Final Clinical Trails Data.zip", "Final Clinical Trails Data"),
+    ("Final Patients Synthea Data.zip", "Final Patients Synthea Data"),
+]
+
+
+def _ensure_dataset(zip_name: str, folder_name: str) -> None:
+    folder_path = BASE_DIR / folder_name
+    if folder_path.exists() and any(folder_path.iterdir()):
+        print(f"[bootstrap] Found {folder_name}.")
+        return
+
+    print(f"[bootstrap] Downloading {zip_name} from {DATASET_REPO}...")
+    zip_path = hf_hub_download(
+        repo_id=DATASET_REPO,
+        filename=zip_name,
+        repo_type="dataset",
+    )
+
+    print(f"[bootstrap] Extracting {zip_name}...")
+    with zipfile.ZipFile(zip_path, "r") as zf:
+        zf.extractall(BASE_DIR)
+
+    if not folder_path.exists():
+        raise RuntimeError(f"Expected extracted folder not found: {folder_path}")
+
+
+def main() -> None:
+    for zip_name, folder_name in DATASET_ZIPS:
+        _ensure_dataset(zip_name, folder_name)
+
+    from app import app
+
+    port = int(os.environ.get("PORT", 7860))
+    app.run(host="0.0.0.0", port=port, debug=False, use_reloader=False)
+
+
+if __name__ == "__main__":
+    main()

llm.md

@@ -0,0 +1,575 @@
+# Second Life — Project Reference (LLM Context Document)
+DSCI 5260 | Group 7 | Last updated: 2026-04-25 (post Session 6 — inbox, messaging, trial dashboard, document upload, dataset seeding)
+
+## Architecture Overview
+
+Flask web app (port 5000) with two authenticated portals:
+- **Patient Portal** (/patient): Register/login, update medical profile, get AI trial matches, connect with hospitals
+- **Hospital Portal** (/hospital): Login, browse opt-in patients, search by condition, manage connections, edit profile
+
+### Key Files
+- `pipeline.py` — Core ML pipeline (data loading, feature engineering, model training, matching)
+- `app.py` — Flask server: session auth, patient API, hospital API, pipeline API
+- `database.py` — SQLite layer: 5 tables, auth, connections, trial interests, messages
+- `templates/landing.html` — Login/register landing page
+- `templates/patient.html` — Patient SPA (profile, trials, connections, inbox)
+- `templates/hospital.html` — Hospital SPA (patients, search, my trials, inbox, connections, profile)
+- `llm.md` — This file
+
+## Running the System
+
+```powershell
+cd "E:\DSCI 5260\Project\PT"
+python app.py
+# open http://localhost:5000
+```
+
+On first run (no model_cache.pkl): loads all data files (~2-3 min), trains RF model, saves cache.
+On subsequent runs: loads cached model immediately.
+
+Delete `model_cache.pkl` to force retrain (required after pipeline feature changes).
+
+## Demo Credentials
+
+All hand-made patients have `open_to_trials=1` and password `pass123`.
+
+| Username | Password | Name | Conditions | Location |
+|----------|----------|------|-----------|----------|
+| john_doe | pass123 | John Doe | hypertension, diabetes, MI | Boston, MA |
+| jane_smith | pass123 | Jane Smith | asthma, atopic dermatitis, allergic rhinitis | Cambridge, MA |
+| bob_jones | pass123 | Robert Jones | CAD, hypertension, chronic pain | Cleveland, OH |
+| alice_brown | pass123 | Alice Brown | non-small cell lung cancer, stroke | Baltimore, MD |
+| david_chen | pass123 | David Chen | diabetes, osteoporosis, CAD | Chicago, IL |
+
+| Username | Password | Hospital | Location |
+|----------|----------|---------|----------|
+| mgh | mgh123 | Massachusetts General Hospital | Boston, MA |
+| cleveland | clinic123 | Cleveland Clinic | Cleveland, OH |
+| jhopkins | johns123 | Johns Hopkins Hospital | Baltimore, MD |
+
+### Dataset-Backed Patients (Synthea)
+20 real Synthea patients are auto-seeded on first run from `Final Patients Synthea Data/`. Username format: `synthea_<first 8 chars of Patient_ID>`, password `pass123`, all `open_to_trials=1`. Total DB patients: 25 (5 hand-made + 20 Synthea).
+
+List all accounts: `SELECT username, first_name, last_name FROM patient_accounts ORDER BY username;`
+
+---
+
+## Data Sources
+
+### Patient Side (Synthea synthetic data)
+- `Final Patients Synthea Data/final_patients_conditions.csv` — ~967K rows, **265,893 patients**, 106 conditions overlapping with trials. Columns: Patient_ID, Condition_Name, Condition_End_Date
+- `Final Patients Synthea Data/patients_details.csv` — Demographics. Columns: Patient_ID, First_Name, Last_Name, Birth_Date (DD-MM-YYYY), Gender, Race, Ethnicity, Address
+- `Final Patients Synthea Data/patients_medications.csv` — **213,182 patients with med data**. Columns: Patient_ID, Medication_Name, Medication_End_Date
+- `Final Patients Synthea Data/patients_observations.csv` — **23,231 patients with lab data**. Columns: Patient_ID, Observation_Name
+
+### Trial Side (ClinicalTrials.gov / AACT)
+- `Final Clinical Trails Data/trail_conditions.csv` — ~1M rows, **571,379 total trials**, **34,074 with matched condition profiles**. Columns: Trial_ID, Condition_Name_Lower
+- `Final Clinical Trails Data/trail_eligibilities.csv` — Columns: Trial_ID, Gender (leading space — stripped), Minimum_Age, Maximum_Age, Eligibility_Criteria
+- `Final Clinical Trails Data/trail_studies.csv` — **65,292 recruiting trials**. Columns: Trial_ID, Brief_Title, Overall_Status, Phase, Start_Date, Enrollment
+- `Final Clinical Trails Data/trail_facilities.csv` — **189,274 trials with US state geo data**. Columns: Trial_ID, Facility_Name, Facility_City, Facility_State (full names), Facility_Country
+- `Final Clinical Trails Data/trail_brief_summaries.csv` — Columns: Trial_ID, Brief_Summary
+- `Final Clinical Trails Data/trail_interventions.csv` — **196,865 trials with drug intervention data**. Columns: Trial_ID, Intervention_Type, Intervention_Name
+- `Final Clinical Trails Data/trail_countries.csv` — Not used for geo scoring (superseded by facility-level state data)
+- `Final Clinical Trails Data/trail_keywords.csv` — Columns: Trial_ID, Keyword_Name_Lower
+
+### MIMIC-IV Demo (code-level validation only, not in UI)
+- `mimic-iv-clinical-database-demo-2.2/hosp/patients.csv.gz`
+- `mimic-iv-clinical-database-demo-2.2/hosp/diagnoses_icd.csv.gz`
+- `mimic-iv-clinical-database-demo-2.2/hosp/d_icd_diagnoses.csv.gz`
+
+---
+
+## SQLite Database (secondlife.db)
+
+### Tables
+```sql
+patient_accounts(id, username, password_hash, synthea_id, first_name, last_name,
+                 dob, gender, address, conditions TEXT DEFAULT '[]',
+                 medications TEXT DEFAULT '[]', documents TEXT DEFAULT '[]',
+                 open_to_trials INTEGER DEFAULT 0, created_at)
+
+hospital_accounts(id, username, password_hash, hospital_name, location,
+                  research_conditions TEXT DEFAULT '[]', created_at)
+
+patient_trial_interests(id, patient_id, trial_id, trial_title, match_score,
+                        status DEFAULT 'interested', created_at,
+                        UNIQUE(patient_id, trial_id))
+
+connections(id, patient_id, hospital_id, trial_id, trial_title,
+            initiated_by DEFAULT 'patient', status DEFAULT 'pending',
+            message, created_at,
+            UNIQUE(patient_id, hospital_id, trial_id))
+-- Post-schema index handles NULL trial_id:
+-- CREATE UNIQUE INDEX idx_conn_unique ON connections(patient_id, hospital_id, COALESCE(trial_id, ''))
+
+connection_messages(id, connection_id, sender_role TEXT,  -- 'patient' or 'hospital'
+                    sender_id TEXT, body TEXT,
+                    created_at, is_read INTEGER DEFAULT 0)
+-- Index: idx_msgs_conn ON connection_messages(connection_id, created_at)
+```
+
+JSON fields (conditions, medications, documents, research_conditions) are stored as TEXT and parsed via `_row_to_dict()`.
+
+### Key database.py Functions
+
+| Function | Purpose |
+|---------|---------|
+| `get_open_patients_for_hospital(hid, condition_filter, include_connected)` | Returns open_to_trials=1 patients; when `include_connected=False` excludes already-connected patients |
+| `get_connection_messages(connection_id)` | All messages for a connection, ASC order |
+| `create_connection_message(connection_id, sender_role, sender_id, body)` | Insert message, returns dict |
+| `mark_messages_read(connection_id, reader_role)` | Mark all messages from the other role as read |
+| `unread_count(connection_id, reader_role)` | Count of unread messages from the other role |
+| `get_hospital_inbox_threads(hospital_id)` | All threads with last_message, last_message_at, unread_count; sorted by activity |
+| `get_patient_inbox_threads(patient_id)` | Same for patient side |
+| `_seed_dataset_patients(c, max_patients=20)` | Seeds Synthea patients from CSV on first run |
+
+---
+
+## Feature Engineering (17 features in FEATURE_COLS)
+
+| Feature | Description |
+|---------|-------------|
+| condition_overlap | Raw count of shared conditions |
+| jaccard_similarity | overlap / union |
+| overlap_ratio_trial | overlap / len(trial_conditions) |
+| overlap_ratio_patient | overlap / len(patient_conditions) |
+| condition_rarity_score | mean(1/log2(n_trials_per_cond+2)), normalised 0-1 |
+| trial_specificity | 1 / trial_condition_count |
+| condition_burden | total_patient_conds / 10 |
+| active_ratio | active_conds / total_conds |
+| resolved_ratio | resolved_conds / total_conds |
+| age_distance | normalised distance outside age range (0 if within) |
+| age_centered | position within age range (-1 to +1) |
+| age_compatibility | 1.0 in range, decays over 30-year gap |
+| gender_compatibility | 1.0 match/all, 0.1 mismatch |
+| **geo_feasibility** | **State-level: 1.0 same state, 0.75 other US state, 0.5 no data** |
+| **med_compatibility** | **Keyword overlap: patient meds vs trial drug interventions** |
+| **lab_availability** | **Patient observation/lab type coverage (0-1, normalised by 20)** |
+| data_completeness | fraction of key fields present |
+
+---
+
+## Flask API Routes
+
+### Auth (no login required)
+- `POST /auth/patient/register` → {success, patient} or {error}
+- `POST /auth/patient/login` → {success, patient} or {error}
+- `POST /auth/hospital/register` → {success, hospital} or {error}
+- `POST /auth/hospital/login` → {success, hospital} or {error}
+- `POST /auth/logout` → {success}
+
+### Patient API (requires patient session)
+- `GET  /api/patient/profile` → patient dict (no password_hash)
+- `POST /api/patient/profile` → updated patient dict; allowed fields: first_name, last_name, dob, gender, address, conditions, medications, open_to_trials
+- `GET  /api/patient/matches` → {results: [...trials], total}
+- `GET  /api/patient/interests` → {interests: [...]}
+- `POST /api/patient/interest` → {success}; body: {trial_id, trial_title, match_score}
+- `DELETE /api/patient/interest/<trial_id>` → {success}
+- `GET  /api/patient/connections` → {connections: [...]} joined with hospital_name
+- `POST /api/patient/connect` → {success, connection} or 409 if duplicate; body: {hospital_id, trial_id, trial_title, message}
+- `GET  /api/patient/hospitals-for-trial?trial_id=NCT...` → {hospitals: [...]} tiered matching (see below)
+- `GET  /api/patient/connections/<cid>/messages` → {messages: [...]}; marks hospital messages read
+- `POST /api/patient/connections/<cid>/messages` → {success, message}; body: {body}
+- `GET  /api/patient/inbox` → {threads: [...]} each with last_message, last_message_at, unread_count, hospital_name
+- `GET  /api/patient/documents` → {documents: [...]}
+- `POST /api/patient/documents` → {success, document}; multipart/form-data file upload (max 10 MB, .pdf/.docx/.doc/.txt/.png/.jpg/.jpeg)
+- `DELETE /api/patient/documents/<doc_id>` → {success}; removes file from disk and DB
+- `GET  /api/patient/documents/<doc_id>/download` → file download (as_attachment)
+
+### Hospital API (requires hospital session)
+- `GET  /api/hospital/profile` → hospital dict (no password_hash)
+- `POST /api/hospital/profile` → updated hospital dict; allowed fields: hospital_name, location, research_conditions
+- `GET  /api/hospital/patients?condition=&include_connected=` → {patients: [...]} open_to_trials=1; `include_connected=true` to include already-connected patients (used by Search tab)
+- `POST /api/hospital/connect` → {success, connection} or 409 if duplicate; body: {patient_id, trial_id, trial_title, message}
+- `GET  /api/hospital/connections` → {connections: [...]} joined with patient fields
+- `PUT  /api/hospital/connections/<cid>/status` → {success}; body: {status: pending|accepted|rejected|completed}
+- `GET  /api/hospital/connections/<cid>/messages` → {messages: [...]}; marks patient messages read
+- `POST /api/hospital/connections/<cid>/messages` → {success, message}; body: {body}
+- `GET  /api/hospital/inbox` → {threads: [...]} each with last_message, last_message_at, unread_count, first_name, last_name
+- `GET  /api/hospital/trials` → {trials: [...]} active trials matched to hospital profile (see below)
+
+### Shared
+- `GET /api/status` → {ready, stats} or {ready: false, message}
+- `GET /api/conditions/autocomplete?q=...` → {results: [...]}
+
+---
+
+## Hospital Trial Dashboard (`/api/hospital/trials`)
+
+`pipeline.trials_for_hospital(hospital_name, location, research_conditions, top_k=20)` — reverse of patient matching: given a hospital's profile, find active clinical trials it is most relevant to.
+
+| Tier | Match condition | `match_reason` field |
+|------|----------------|----------------------|
+| 1 | Jaccard(hospital name tokens, trial facility name tokens) ≥ 0.25 | "name matched to trial site" |
+| 2 | Hospital state matches a US trial facility state | "in same state as trial site" |
+| 3 | Hospital `research_conditions` overlaps trial conditions | "researches related conditions" |
+
+Active-only filter (`is_active` check) applied at every tier. Returns list of dicts:
+`trial_id, title, phase, status, summary, location, facility_name, n_sites, match_tier, match_reason`
+
+---
+
+## Tiered Hospital Matching (`/api/patient/hospitals-for-trial`)
+
+For each trial, hospitals in the DB are scored and returned in tier order (Tier 1 first):
+
+| Tier | Match condition | `match_reason` field | UI label |
+|------|----------------|----------------------|----------|
+| 1 | Jaccard(hospital name tokens, any trial facility name tokens) ≥ 0.25 | "verified site on this trial" | Green — Verified Trial Sites |
+| 2 | Hospital state (from "City, ST" location) matches a trial US facility state | "in same state as a trial site" | Grey — Related Hospitals |
+| 3 | Hospital `research_conditions` overlaps trial conditions | "researches related conditions" | Grey — Related Hospitals |
+| 4 | Fallback — trial has no facility/state data at all | "" | Grey — Related Hospitals |
+
+The patient connect modal groups Tier 1 hospitals under a green "VERIFIED TRIAL SITES" header and Tiers 2-4 under a grey "RELATED HOSPITALS — not confirmed trial sites" header.
+
+**Pipeline lookups used:**
+- `pipeline.trial_facility_tokens[trial_id]` — list of frozensets of significant words from facility names
+- `pipeline.trial_us_states[trial_id]` — set of US state full names (e.g. {"Massachusetts"})
+- `pipeline.trial_profiles[trial_id]["conditions"]` — set of condition strings
+
+**Stopword set for facility name tokenisation** (same in pipeline.py and app.py):
+hospital, medical, center, centre, clinic, university, health, care, healthcare, system, institute, foundation, research, general, regional, national, community, services, department, division, college, school, the, of, and, at, for, in, a, an, is, by
+
+---
+
+## Trial Match Result Fields
+
+Each item in `/api/patient/matches` results:
+- trial_id, title, phase, status, min_age, max_age, sex, enrollment, start_date
+- eligibility_probability (0-100, calibrated RF probability × 100)
+- match_score (0-100, rule-based: overlap_ratio weighted)
+- combined_score (0-100, 0.6 × eligibility + 0.4 × match_score)
+- age_compatibility, gender_compatibility, geo_feasibility, med_compatibility (all 0-100)
+- condition_rarity_score (0-1)
+- overlap_conditions (list of conditions shared with patient)
+- trial_conditions (all trial conditions)
+- criteria (eligibility criteria text, truncated 500 chars)
+- summary (brief summary, truncated 400 chars)
+- **facility_name** (lead US facility name, or "" if not available)
+- location (lead US facility city/state/country string)
+- n_sites (total facility count for this trial)
+- interest_status (null | 'interested' | 'withdrawn', from patient_trial_interests)
+
+---
+
+## Data Privacy Model
+
+1. Hospitals see only patients with open_to_trials=1 (name, age, gender, conditions)
+2. Full details accessible only after patient-initiated connection
+3. Hospital cannot contact a patient unless patient is open to trials
+4. Connection record: patient_id, hospital_id, trial_id, initiated_by, status, message
+5. Hospital can also initiate connections with opt-in patients from the hospital portal
+
+---
+
+## XSS Prevention
+
+All user-controlled strings use DOM API (never innerHTML for user data):
+```javascript
+function escH(s) {       // text content in innerHTML contexts
+  const d = document.createElement('div');
+  d.appendChild(document.createTextNode(String(s||'')));
+  return d.innerHTML;
+}
+function escA(s) {       // HTML attribute values
+  return String(s||'').replace(/&/g,'&amp;').replace(/"/g,'&quot;')
+                      .replace(/</g,'&lt;').replace(/>/g,'&gt;');
+}
+```
+Event listeners use addEventListener only. Tags and cards built via createElement + textContent.
+
+---
+
+## ML Model
+
+- Random Forest (n_estimators=200, max_depth=12, class_weight="balanced")
+- CalibratedClassifierCV (isotonic, cv=3) for probability calibration
+- GroupShuffleSplit (patient-level, 80/20, no leakage) for train/test split
+- GroupKFold (5-fold, patient-level) for cross-validation
+- Training sample: 3000 patients × 30 trials each + random negatives
+- Cache: `model_cache.pkl` (delete to force retrain)
+
+### Actual Metrics (from verified live run, 2026-04-25)
+
+| Metric | Value |
+|--------|-------|
+| Accuracy | 85.15% |
+| AUC-ROC | 0.5976 |
+| CV AUC (5-fold) | 0.5992 ± 0.0052 |
+| F1 | 0.9168 |
+| Precision | 0.8518 |
+| Recall | 0.9925 |
+| Brier score | 0.1263 |
+| Avg precision | 0.8540 |
+| Train size | 90,994 pairs |
+| Test size | 22,681 pairs |
+| Positive label rate | 82.1% |
+
+> **Note on AUC:** The 82.1% positive rate in pseudo-labels (weighted 6-feature labelling threshold at 0.5) makes the classification task easy to solve trivially — high accuracy/recall but lower AUC. To improve AUC, the pseudo-label threshold should be raised (e.g. 0.6) or positive/negative sampling balanced more aggressively.
+
+### Feature Importance (Random Forest, ranked)
+
+| Rank | Feature | Importance |
+|------|---------|-----------|
+| 1 | age_distance | 0.2040 |
+| 2 | age_compatibility | 0.1691 |
+| 3 | gender_compatibility | 0.1288 |
+| 4 | age_centered | 0.0791 |
+| 5 | jaccard_similarity | 0.0751 |
+| 6 | condition_rarity_score | 0.0621 |
+| 7 | overlap_ratio_trial | 0.0519 |
+| 8 | overlap_ratio_patient | 0.0512 |
+| 9 | condition_overlap | 0.0396 |
+| 10 | condition_burden | 0.0274 |
+| 11 | resolved_ratio | 0.0263 |
+| 12 | active_ratio | 0.0258 |
+| 13 | lab_availability | 0.0208 |
+| 14 | geo_feasibility | 0.0158 |
+| 15 | med_compatibility | 0.0144 |
+| 16 | trial_specificity | 0.0087 |
+| 17 | data_completeness | 0.0000 |
+
+---
+
+## MIMIC-IV Validation (code-level only, not in UI)
+
+- 100 demo patients, ~90% match rate after 3-tier ICD → condition mapping
+- Call: `pipeline.validate_mimic()` → list of {subject_id, mapped_conditions, n_matches, top_match}
+- 3-tier mapping: exact → substring containment → word-overlap ≥ 75%
+- Not exposed via any Flask route
+
+---
+
+## All Bug Fixes by Session
+
+### Session 3 Fixes (2026-04-25) — Two-portal foundation
+
+#### Fix 1 — patient_id not passed to match_patient() (CRITICAL)
+**Before:** `pipeline.match_patient(conditions, age, gender, top_k=20)`
+**After:** `pipeline.match_patient(conditions, age, gender, top_k=20, patient_id=session["patient_id"], address=address)`
+
+Without this, patient-specific medication keywords and lab scores defaulted to empty / 0.3 for all users — med_compatibility and lab_availability were effectively constants.
+
+#### Fix 2 — Pseudo-label used only 3 features (HIGH)
+**Before:** AND gate on age/gender/condition — geo/med/lab had near-zero training influence.
+**After:** 6-feature weighted score with 15% random noise:
+```python
+score = (
+    0.30 * float(row["age_compatibility"]    > 0.6) +
+    0.15 * float(row["gender_compatibility"] > 0.5) +
+    0.25 * float(row["jaccard_similarity"]   > 0.05) +
+    0.10 * float(row["geo_feasibility"]) +
+    0.10 * float(row["med_compatibility"]) +
+    0.10 * float(row["lab_availability"])
+)
+base = int(score >= 0.5)
+# 15% hash-deterministic noise for realism
+```
+
+#### Fix 3 — geo_feasibility was country-level heuristic (MEDIUM)
+**Before:** Float from `trail_countries.csv` (1.0 US, 0.7 multi-national, 0.35 non-US). No patient location.
+**After:** State-level matching using `trail_facilities.csv` + patient address regex:
+```python
+def _geo_score(patient_state_full, trial_states):
+    if not trial_states: return 0.5       # no US facility data — neutral
+    if patient_state_full in trial_states: return 1.0
+    return 0.75                           # other US state
+```
+
+#### Fix 4 — NameError `trial_geo` in _compute_features return dict
+`"geo_feasibility": float(trial_geo)` → `"geo_feasibility": geo_feasibility`
+
+#### Fix 5 — XSS in condition tag onclick handlers (MEDIUM)
+`addConditionTag('${c}')` broke for conditions with apostrophes (e.g. "alzheimer's disease").
+Fixed with DOM-based `makeTag()` using textContent + addEventListener. No inline onclick anywhere.
+
+#### Fix 6 — Duplicate connection prevention (was: no guard)
+- `connections` table: added `UNIQUE(patient_id, hospital_id, trial_id)` schema constraint
+- `init_db()`: runs `CREATE UNIQUE INDEX IF NOT EXISTS idx_conn_unique ON connections(patient_id, hospital_id, COALESCE(trial_id, ''))` to handle NULL trial_id and backfill existing DBs
+- `create_connection()`: pre-checks `connection_exists()` before insert; returns `None` on duplicate
+- `/api/patient/connect` and `/api/hospital/connect`: return 409 when `create_connection()` returns None
+
+#### Fix 7 — Hospital patient feed showed already-contacted patients (was: no exclusion)
+`get_open_patients_for_hospital()` now uses:
+```sql
+WHERE open_to_trials=1
+  AND id NOT IN (SELECT DISTINCT patient_id FROM connections WHERE hospital_id=?)
+```
+
+#### Fix 8 — Demo seed: john_doe starts with open_to_trials=1
+Hospital portal was empty on a fresh database. `_seed_demo_data()` now seeds john_doe with `open_to_trials=1`.
+
+#### Fix 9 — Hospital registration silently ignored research_conditions
+`templates/landing.html` hospital register form now collects comma-separated research conditions and sends them as a parsed lowercase array to the backend.
+
+#### Fix 10 — Trial cards only showed site count, not facility name or location
+`pipeline.py match_patient()` now extracts `facility_name` from `Facility_Name` column; prefers US facilities. Patient portal detail grid shows "Lead Site" and "Location" when available.
+
+---
+
+### Session 4 Fixes (2026-04-25) — Hospital matching overhaul + profile editing
+
+#### Fix 11 — Hospital suggestion logic replaced (was: research_conditions overlap only)
+Complete replacement of `/api/patient/hospitals-for-trial`:
+
+**Before:** looped all hospitals, included any whose `research_conditions` overlapped trial conditions. No tier concept, no facility data used.
+
+**After:** 4-tier system using two new pipeline lookups:
+- `pipeline.trial_facility_tokens[trial_id]` — built from `Facility_Name` column in trail_facilities.csv, US rows only. Each facility name tokenised by stripping stopwords + words < 3 chars.
+- `pipeline.trial_us_states[trial_id]` — set of full US state names for the trial
+
+Helper functions in `app.py`:
+```python
+def _hospital_name_tokens(name: str) -> frozenset:
+    # strips stopwords, keeps words ≥ 3 chars
+    ...
+
+def _facility_match_score(h_tokens: frozenset, facility_token_list: list) -> float:
+    # best Jaccard score against any facility in the trial
+    ...
+```
+
+Each hospital gets one tier assigned and a `match_reason` + `match_tier` in the response.
+Result list sorted by `match_tier` ascending (best first).
+
+#### Fix 12 — Hospital portal had no profile editing
+`POST /api/hospital/profile` added (was GET-only). `database.py update_hospital_profile()` added. `templates/hospital.html` now has a **My Profile** tab with editable hospital name, location, and research condition tags. On save, the navbar hospital name updates live without a page reload.
+
+#### Fix 13 — Model disclaimer missing from patient trial results
+`templates/patient.html` trial results section now shows an alert above results:
+> "Match percentages are predictions from a model trained on synthetic patient data and rule-based labels — not validated clinical eligibility determinations. Always consult a healthcare provider before enrolling in any trial."
+
+---
+
+### Session 5 Fixes (2026-04-25) — Modal UX + bug fixes
+
+#### Fix 14 — Patient connect modal showed all hospitals in one flat list
+**Before:** All hospitals (all tiers) in a single flat list, sorted by tier, with coloured badges as the only visual distinction.
+
+**After:** Modal renders two visually separated sections:
+- **"VERIFIED TRIAL SITES"** (green `sec-head`) — Tier 1 hospitals only
+- **"RELATED HOSPITALS — not confirmed trial sites"** (grey `sec-head` with inline subtitle) — Tiers 2, 3, 4
+
+Both sections only render if they have entries. Click delegation on the outer `#hospitalList` wrapper still works for both sections.
+
+#### Fix 15 — Close button invisible on hospital profile condition tags
+`hospital.html renderProfTags()`: `btn-close-white` (white X) on `bg-info text-dark` badge (light blue background) → `btn-close` (dark X). The X was invisible before.
+
+#### Fix 16 — Login/register forms required mouse click, no Enter key support
+`templates/landing.html`: Added `_onEnter(inputId, fn)` helper and wired Enter key on all login and register inputs (both patient and hospital portals). Works on username field too (not just password).
+
+#### Fix 17 — System ready banner never auto-cleared on slow boot
+`templates/patient.html`: `checkStatus()` was called once at DOMContentLoaded and never again. If the pipeline was still training when the user opened the page, the yellow banner persisted even after the pipeline finished.
+
+**After:** `startStatusPoll()` starts a `setInterval` (5s) when the initial check finds `ready: false`. The interval clears itself once `ready: true` is received.
+```javascript
+checkStatus().then(() => { if (!sysReady) startStatusPoll(); });
+```
+
+---
+
+## Live Run Verification (2026-04-25)
+
+End-to-end test results after full retrain with no model_cache.pkl:
+
+| Test | Result |
+|------|--------|
+| Landing page GET / | 200 OK |
+| Patient login john_doe/pass123 | OK — returns patient JSON |
+| Hospital login mgh/mgh123 | OK — returns hospital JSON |
+| Pipeline ready (api/status) | ready: true |
+| /api/patient/matches for john_doe | 20 results, all score fields populated |
+| Top match geo score for Greece trial | 50% (no US facility — correct) |
+| Hospital browses open_to_trials patients | 25 patients visible (5 demo + 20 Synthea) |
+| Hospital condition search ?condition=hypertension | results including Synthea patients |
+| Hospital → patient connect (POST) | OK, status=pending |
+| Patient sees hospital connection (GET) | 1 connection, hospital_name present |
+| /api/conditions/autocomplete?q=hyper | ["hypertension"] |
+| Duplicate connect attempt | 409 error |
+| Hospital profile save | navbar name updates live |
+| Patient connect modal | Two sections render correctly |
+| /api/hospital/trials for mgh | Active trials with match tiers |
+| /api/hospital/inbox | Threads with unread counts |
+| /api/patient/inbox | Threads with hospital names |
+| Patient document upload | File saved, metadata in DB |
+| Inbox Synthea seeding | 25 patients total confirmed |
+
+---
+
+### Session 6 Fixes (2026-04-25) — Messaging, inbox, trial dashboard, document upload, dataset seeding
+
+#### Fix 18 — Hospital trial dashboard (My Trials tab)
+**Before:** Hospital portal had no way to see which clinical trials were relevant to it.
+
+**After:** New "My Trials" nav tab in `hospital.html`. Calls `GET /api/hospital/trials` → `pipeline.trials_for_hospital()`. Active-only filter at all 3 tiers. Cards show status badge, phase, match tier, facility name, summary excerpt, and a "View on ClinicalTrials.gov ↗" link.
+
+Active-only filter: `if not self.trial_profiles[trial_id].get("is_active", False): continue` at each tier loop in `pipeline.py`.
+
+#### Fix 19 — Messaging (chat in connections)
+**Before:** Connections table had no messaging. Patients and hospitals could only see connection status.
+
+**After:**
+- New `connection_messages` table with `sender_role`, `sender_id`, `body`, `is_read`.
+- `GET/POST /api/patient/connections/<cid>/messages` and `GET/POST /api/hospital/connections/<cid>/messages`.
+- Both portals have a messages modal (`#msgModal`) opened by a Chat button in the Connections table.
+- `mark_messages_read()` called on GET to auto-mark messages as read when the recipient opens the thread.
+
+#### Fix 20 — Dedicated Inbox tab (both portals)
+**Before:** Chat only accessible from the My Connections table row — no inbox overview.
+
+**After:** New "Inbox" nav tab in both `hospital.html` and `patient.html`.
+- Calls `GET /api/hospital/inbox` or `GET /api/patient/inbox`.
+- Backed by `get_hospital_inbox_threads()` / `get_patient_inbox_threads()` — SQL subqueries aggregate last_message, last_message_at, unread_count per thread.
+- Threads sorted by most recent activity (Python-side sort on `last_message_at or created_at`).
+- Unread count badge on nav tab button updates when inbox loads.
+- "Open" button reuses the existing `openMsgModal()` and messages modal.
+
+#### Fix 21 — Document upload (patient portal)
+**Before:** Patient profile had no file upload section.
+
+**After:** "My Documents" card added to patient profile tab. 4 routes:
+- `POST /api/patient/documents` — werkzeug `secure_filename`, 10 MB limit, allowed extensions: `.pdf/.docx/.doc/.txt/.png/.jpg/.jpeg`. Saves to `uploads/patient_docs/<patient_id>/`. Metadata stored as JSON array in `patient_accounts.documents`.
+- `DELETE /api/patient/documents/<doc_id>` — removes file from disk and metadata from DB.
+- `GET /api/patient/documents/<doc_id>/download` — serves file as attachment.
+- `app.config["MAX_CONTENT_LENGTH"] = 10 * 1024 * 1024` enforced Flask-side.
+
+#### Fix 22 — Dataset-backed patient seeding
+**Before:** Only hand-made demo patients in DB (john_doe only had open_to_trials=1 initially). Hospital search returned 0 results on fresh DB.
+
+**After:** `_seed_dataset_patients(c, max_patients=20)` in `database.py` seeds 20 real Synthea patients from CSV files on first run. Skips deceased patients (Death_Date not empty). Reads up to 8 conditions + 6 medications per patient. Username: `synthea_<first8chars_of_Patient_ID>`, password `pass123`, `open_to_trials=1`.
+
+All 5 hand-made demo patients also set to `open_to_trials=1`. Total: 25 patients in DB.
+
+#### Fix 23 — Hospital search include_connected toggle
+**Before:** Hospital Search tab also excluded already-connected patients, same as Available Patients tab — making it useless for re-searching.
+
+**After:** Search tab adds `include_connected=true` query param. `GET /api/hospital/patients?include_connected=true` bypasses the exclusion subquery. Available Patients tab retains strict exclusion. Toggle checkbox in Search tab UI.
+
+#### Fix 24 — Template auto-reload
+**Before:** `app.run(debug=False, use_reloader=False)` — template edits required server restart to take effect.
+
+**After:** `app.config["TEMPLATES_AUTO_RELOAD"] = True` added after other config lines. Templates now reload on every request without enabling full debug mode or the reloader.
+
+---
+
+## Known Issues / Future Improvements
+
+1. **High pseudo-label positive rate (82.1%)** — lowers AUC-ROC to ~0.60. Fix: raise label threshold from 0.5 to 0.6, or explicitly sample equal positive/negative pairs.
+
+2. **data_completeness feature importance = 0** — nearly constant across training pairs (all synthetic patients have complete data). Consider removing from FEATURE_COLS.
+
+3. **Fuzzy condition matching not implemented** — only exact condition name overlaps used (106 conditions). Substring/semantic fuzzy matching would expand coverage significantly.
+
+4. **lab_availability coverage is low (8.7%)** — observations file is sparse. Consider normalising denominator to the subset that has any lab data.
+
+5. **Hospital portal does not rank patients by match quality** — listed in insertion order. Could rank by condition overlap with the hospital's research_conditions.
+
+6. **No email/notification system** — connection requests visible only inside the portal.
+
+7. **No automated tests** — syntax checking only (`python -m py_compile`). Key flows to cover: patient registration → profile update → match → connect; hospital registration → patient browse → connect → status update; duplicate connection rejection.
+
+8. **Inbox badge not auto-refreshed** — unread count badge only updates when the user clicks the Inbox tab. No real-time push; would require polling or WebSockets.
+
+9. **Document access control** — uploaded files are served from disk by doc_id only; no additional hospital-side access to patient documents (by design — privacy model). Hospital sees document count in patient profile only after connection.
+
+10. **Synthea patients have Synthea-style names** (e.g. "Geovany567 Reichert456") — cosmetically odd but functionally correct. No fix needed for demo.

login.md

@@ -0,0 +1,38 @@
+# Demo Login Credentials
+
+## Hospital Accounts
+
+| Portal | Username | Password | Notes |
+|---|---|---|---|
+| Hospital | `mgh` | `mgh123` | Massachusetts General Hospital |
+| Hospital | `cleveland` | `clinic123` | Cleveland Clinic |
+| Hospital | `jhopkins` | `johns123` | Johns Hopkins Hospital |
+
+## Demo Patient Accounts
+
+| Portal | Username | Password | Notes |
+|---|---|---|---|
+| Patient | `john_doe` | `pass123` | Demo patient |
+| Patient | `jane_smith` | `pass123` | Demo patient |
+| Patient | `bob_jones` | `pass123` | Demo patient |
+| Patient | `alice_brown` | `pass123` | Demo patient |
+| Patient | `david_chen` | `pass123` | Demo patient |
+
+## Dataset-Backed Patient Accounts
+
+- The app also auto-seeds a small set of Synthea patients into the portal.
+- Username format: `synthea_<first 8 chars of Patient_ID>`
+- Default password: `pass123`
+- Example: patient ID `660bec03-...` becomes username `synthea_660bec03`
+
+To list all current patient usernames in the local DB:
+
+```powershell
+@'
+import sqlite3
+con = sqlite3.connect("secondlife.db")
+cur = con.cursor()
+for row in cur.execute("select username, first_name, last_name, synthea_id from patient_accounts order by username"):
+    print(row)
+'@ | python -
+```

model_cache.pkl

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b8e3c50d92c2b71314f765dba7d4cacb1f7a349ed4a9d0153b222bf48cf14e83
+size 79928619

pipeline.py

@@ -0,0 +1,1153 @@
+"""
+Second Life — Clinical Trial Matching Pipeline
+DSCI 5260 | Group 7
+
+Full pipeline using ALL available data files:
+  Patient side  : conditions, demographics, medications, observations
+  Trial side    : conditions, eligibilities, studies, facilities,
+                  summaries, interventions, countries, keywords
+  Validation    : MIMIC-IV (real patients, NDA access)
+
+Features align with Second_Life_Final_Notebook_V2.ipynb plus
+real-data replacements for previously simulated feasibility factors.
+"""
+
+import re
+import math
+import pickle
+import warnings
+from pathlib import Path
+from datetime import datetime
+
+# US state abbreviation → full name (patient addresses use abbreviations,
+# facility data uses full names)
+STATE_ABBREV = {
+    "AL":"Alabama","AK":"Alaska","AZ":"Arizona","AR":"Arkansas","CA":"California",
+    "CO":"Colorado","CT":"Connecticut","DE":"Delaware","FL":"Florida","GA":"Georgia",
+    "HI":"Hawaii","ID":"Idaho","IL":"Illinois","IN":"Indiana","IA":"Iowa",
+    "KS":"Kansas","KY":"Kentucky","LA":"Louisiana","ME":"Maine","MD":"Maryland",
+    "MA":"Massachusetts","MI":"Michigan","MN":"Minnesota","MS":"Mississippi",
+    "MO":"Missouri","MT":"Montana","NE":"Nebraska","NV":"Nevada","NH":"New Hampshire",
+    "NJ":"New Jersey","NM":"New Mexico","NY":"New York","NC":"North Carolina",
+    "ND":"North Dakota","OH":"Ohio","OK":"Oklahoma","OR":"Oregon","PA":"Pennsylvania",
+    "RI":"Rhode Island","SC":"South Carolina","SD":"South Dakota","TN":"Tennessee",
+    "TX":"Texas","UT":"Utah","VT":"Vermont","VA":"Virginia","WA":"Washington",
+    "WV":"West Virginia","WI":"Wisconsin","WY":"Wyoming","DC":"District of Columbia",
+}
+
+import numpy as np
+import pandas as pd
+from sklearn.calibration import CalibratedClassifierCV
+from sklearn.ensemble import RandomForestClassifier
+from sklearn.metrics import (
+    accuracy_score, average_precision_score, brier_score_loss,
+    f1_score, precision_score, recall_score, roc_auc_score,
+)
+from sklearn.model_selection import GroupKFold, GroupShuffleSplit, cross_val_score
+from sklearn.preprocessing import StandardScaler
+
+warnings.filterwarnings("ignore")
+
+# ---------------------------------------------------------------------------
+# Paths
+# ---------------------------------------------------------------------------
+BASE_DIR = Path(__file__).parent
+
+# Patient data (Synthea)
+PATIENT_CONDITIONS_PATH  = BASE_DIR / "Final Patients Synthea Data" / "final_patients_conditions.csv"
+PATIENT_DETAILS_PATH     = BASE_DIR / "Final Patients Synthea Data" / "patients_details.csv"
+PATIENT_MEDICATIONS_PATH = BASE_DIR / "Final Patients Synthea Data" / "patients_medications.csv"
+PATIENT_OBSERVATIONS_PATH= BASE_DIR / "Final Patients Synthea Data" / "patients_observations.csv"
+
+# Trial data (ClinicalTrials.gov / AACT)
+TRIAL_CONDITIONS_PATH    = BASE_DIR / "Final Clinical Trails Data" / "trail_conditions.csv"
+TRIAL_ELIGIBILITY_PATH   = BASE_DIR / "Final Clinical Trails Data" / "trail_eligibilities.csv"
+TRIAL_STUDIES_PATH       = BASE_DIR / "Final Clinical Trails Data" / "trail_studies.csv"
+TRIAL_FACILITIES_PATH    = BASE_DIR / "Final Clinical Trails Data" / "trail_facilities.csv"
+TRIAL_SUMMARIES_PATH     = BASE_DIR / "Final Clinical Trails Data" / "trail_brief_summaries.csv"
+TRIAL_INTERVENTIONS_PATH = BASE_DIR / "Final Clinical Trails Data" / "trail_interventions.csv"
+TRIAL_COUNTRIES_PATH     = BASE_DIR / "Final Clinical Trails Data" / "trail_countries.csv"
+TRIAL_KEYWORDS_PATH      = BASE_DIR / "Final Clinical Trails Data" / "trail_keywords.csv"
+
+# MIMIC-IV (real patients, NDA)
+MIMIC_PATIENTS_PATH  = BASE_DIR / "mimic-iv-clinical-database-demo-2.2" / "hosp" / "patients.csv.gz"
+MIMIC_DIAGNOSES_PATH = BASE_DIR / "mimic-iv-clinical-database-demo-2.2" / "hosp" / "diagnoses_icd.csv.gz"
+MIMIC_ICD_DICT_PATH  = BASE_DIR / "mimic-iv-clinical-database-demo-2.2" / "hosp" / "d_icd_diagnoses.csv.gz"
+
+MODEL_CACHE = BASE_DIR / "model_cache.pkl"
+
+# ---------------------------------------------------------------------------
+# Feature columns — matches Second_Life_Final_Notebook_V2.ipynb feature set
+# plus real-data replacements for the Beta-simulated feasibility factors
+# ---------------------------------------------------------------------------
+FEATURE_COLS = [
+    # --- Condition matching (from notebook) ---
+    "condition_overlap",          # raw count of shared conditions
+    "jaccard_similarity",         # overlap / union
+    "overlap_ratio_trial",        # overlap / len(trial_conditions)
+    "overlap_ratio_patient",      # overlap / len(patient_conditions)
+    "condition_rarity_score",     # mean(1/log2(n_trials_per_cond+2)) — rare conditions = higher score
+    "trial_specificity",          # 1 / trial_condition_count — focused trials score higher
+    # --- Patient profile ---
+    "condition_burden",           # total patient conditions / 10 (normalised)
+    "active_ratio",               # active (unresolved) conditions / total
+    "resolved_ratio",             # resolved conditions / total  ← NEW
+    # --- Age (soft continuous, no hard gate) ---
+    "age_distance",               # normalised distance outside age range (0 if within)
+    "age_centered",               # position within age range (−1 to +1)
+    "age_compatibility",          # 1.0 in range, decays over 30-year gap
+    # --- Gender (soft) ---
+    "gender_compatibility",       # 1.0 match/all, 0.1 mismatch
+    # --- Real feasibility factors (replace Beta simulation) ---
+    "geo_feasibility",            # 1.0 same-state facility, 0.8 US other state, 0.4 international
+    "med_compatibility",          # keyword overlap: patient meds ↔ trial drug interventions
+    "lab_availability",           # patient observation/lab type coverage (0–1)
+    # --- Data quality ---
+    "data_completeness",          # fraction of key fields present
+]
+
+
+# ---------------------------------------------------------------------------
+# Helpers
+# ---------------------------------------------------------------------------
+
+def _parse_age(age_str):
+    if pd.isna(age_str):
+        return None
+    s = str(age_str).lower().strip()
+    if s in ("n/a", "na", "", "none"):
+        return None
+    nums = re.findall(r"(\d+)", s)
+    if not nums:
+        return None
+    age = int(nums[0])
+    if "month" in s:
+        return age / 12.0
+    if "week" in s:
+        return age / 52.0
+    if "day" in s:
+        return age / 365.0
+    return float(age)
+
+
+def _drug_keyword(name: str) -> str:
+    """Extract the primary drug name keyword from a medication/intervention string."""
+    name = str(name).lower().strip()
+    # Remove leading spaces and numeric prefixes like "24 HR "
+    name = re.sub(r"^\d[\d\s]*hr\s+", "", name)
+    # Remove dosage patterns
+    name = re.sub(
+        r"[\d./]+\s*(mg|ml|mcg|iu|units?|%|actuat|day|pack|tablet|injection"
+        r"|oral|capsule|solution|cream|spray|patch|inhaler|mg/ml|mg/actuat)",
+        " ", name
+    )
+    # Split on non-alpha
+    words = [w for w in re.split(r"[\s,\-+/\[\]()]+", name)
+             if len(w) >= 4 and not w.replace(".", "").replace("/", "").isdigit()]
+    if not words:
+        return ""
+    # Skip generic prefixes
+    skip = {"intra", "oral", "drug", "form", "with", "plus", "anti", "solution",
+            "extended", "release", "pack", "spray", "cream", "patch"}
+    for w in words:
+        if w not in skip:
+            return w
+    return words[0]
+
+
+def _name_tokens(name: str) -> frozenset:
+    """Tokenise a hospital/facility name: drop stopwords and short words."""
+    _STOP = {"the","of","and","at","for","in","a","an","is","by",
+             "hospital","medical","center","centre","clinic","university",
+             "health","care","healthcare","system","institute","foundation",
+             "research","general","regional","national","community",
+             "services","department","division","college","school"}
+    words = re.findall(r"[a-z]+", str(name).lower())
+    return frozenset(w for w in words if w not in _STOP and len(w) >= 3)
+
+
+def _geo_score(patient_state_full: str, trial_states: set) -> float:
+    """
+    State-level geo feasibility score.
+    patient_state_full: full state name e.g. 'Massachusetts'
+    trial_states: set of full US state names where trial has facilities
+    """
+    if not trial_states:
+        return 0.5   # no facility data — neutral
+    if patient_state_full and patient_state_full in trial_states:
+        return 1.0   # trial in same state
+    if trial_states:             # trial has US facility but different state
+        return 0.75
+    return 0.35                  # international only (shouldn't reach here)
+
+
+def _extract_state(address: str) -> str:
+    """Extract full US state name from a patient address string."""
+    if not address or address in ("nan", "None", ""):
+        return ""
+    m = re.search(r"\b([A-Z]{2})\s+\d{5}\b", str(address))
+    if m:
+        return STATE_ABBREV.get(m.group(1), "")
+    return ""
+
+
+def _compute_features(
+    patient_conds: set, patient_age: float, patient_gender: str,
+    trial_conds: set, trial_min_age: float, trial_max_age: float,
+    trial_gender: str,
+    total_patient_conds: int, active_patient_conds: int,
+    trial_cond_count: int,
+    # pre-computed per-entity lookups
+    cond_rarity_map: dict,          # condition → rarity_score
+    trial_us_states: set,           # set of US state full-names for this trial
+    pat_med_kws: set,               # patient's medication keywords
+    trial_drug_kws: set,            # trial's drug intervention keywords
+    pat_lab_score: float,           # patient's lab/observation coverage (0–1)
+    patient_state_full: str = "",   # patient's US state full name
+) -> dict:
+    """Compute all 17 soft continuous match features (no hard binary gates)."""
+
+    overlap = len(patient_conds & trial_conds)
+    union   = len(patient_conds | trial_conds)
+
+    jaccard             = overlap / union if union > 0 else 0.0
+    overlap_ratio_pat   = overlap / len(patient_conds)  if len(patient_conds) > 0  else 0.0
+    overlap_ratio_trial = overlap / len(trial_conds)    if len(trial_conds) > 0    else 0.0
+
+    # Condition rarity: rare condition match is stronger signal
+    overlap_conds = patient_conds & trial_conds
+    if overlap_conds:
+        rarity_vals = [cond_rarity_map.get(c, 0.5) for c in overlap_conds]
+        condition_rarity_score = float(np.mean(rarity_vals))
+    else:
+        condition_rarity_score = 0.0
+
+    trial_specificity   = 1.0 / trial_cond_count if trial_cond_count > 0 else 0.0
+    condition_burden    = min(total_patient_conds / 10.0, 1.0)
+    active_ratio        = active_patient_conds / total_patient_conds if total_patient_conds > 0 else 0.0
+    resolved            = max(0, total_patient_conds - active_patient_conds)
+    resolved_ratio      = resolved / total_patient_conds if total_patient_conds > 0 else 0.0
+
+    # Age features (continuous)
+    age     = patient_age if patient_age and not np.isnan(patient_age) else 50.0
+    min_a   = trial_min_age if trial_min_age else 0.0
+    max_a   = trial_max_age if trial_max_age else 120.0
+
+    if age < min_a:
+        age_distance = (min_a - age) / 100.0
+    elif age > max_a:
+        age_distance = (age - max_a) / 100.0
+    else:
+        age_distance = 0.0
+
+    mid    = (min_a + max_a) / 2.0
+    half   = (max_a - min_a) / 2.0 if max_a > min_a else 1.0
+    age_centered = max(-1.0, min(1.0, (age - mid) / half))
+
+    if min_a <= age <= max_a:
+        age_compat = 1.0
+    else:
+        dist = min(abs(age - min_a), abs(age - max_a))
+        age_compat = max(0.0, 1.0 - dist / 30.0)
+
+    # Gender (soft)
+    pg = str(patient_gender).upper().strip()
+    tg = str(trial_gender).upper().strip()
+    if tg in ("ALL", ""):
+        gender_compat = 1.0
+    elif (pg in ("M", "MALE")   and tg in ("M", "MALE")) or \
+         (pg in ("F", "FEMALE") and tg in ("F", "FEMALE")):
+        gender_compat = 1.0
+    else:
+        gender_compat = 0.1
+
+    # Geo feasibility: state-level (same state = 1.0, other US = 0.75, no data = 0.5)
+    geo_feasibility = _geo_score(patient_state_full, trial_us_states)
+
+    # Medication compatibility (real: keyword overlap)
+    if pat_med_kws and trial_drug_kws:
+        m_overlap = len(pat_med_kws & trial_drug_kws)
+        m_union   = len(pat_med_kws | trial_drug_kws)
+        # Presence of any shared keyword = compatible; jaccard for strength
+        med_compat = 0.8 + 0.2 * (m_overlap / m_union) if m_overlap > 0 else 0.4
+    else:
+        med_compat = 0.5   # neutral when data is missing
+
+    # Data completeness
+    data_completeness = sum([
+        not np.isnan(age),
+        pg != "",
+        total_patient_conds > 0,
+        min_a is not None,
+        tg != "",
+    ]) / 5.0
+
+    # Rule-based match score (reference only — NOT a feature, used for labelling)
+    match_score = (
+        0.35 * overlap_ratio_trial +
+        0.25 * overlap_ratio_pat   +
+        0.20 * age_compat          +
+        0.20 * gender_compat
+    ) * 100.0
+
+    return {
+        "condition_overlap":        float(overlap),
+        "jaccard_similarity":       jaccard,
+        "overlap_ratio_trial":      overlap_ratio_trial,
+        "overlap_ratio_patient":    overlap_ratio_pat,
+        "condition_rarity_score":   condition_rarity_score,
+        "trial_specificity":        trial_specificity,
+        "condition_burden":         condition_burden,
+        "active_ratio":             active_ratio,
+        "resolved_ratio":           resolved_ratio,
+        "age_distance":             age_distance,
+        "age_centered":             age_centered,
+        "age_compatibility":        age_compat,
+        "gender_compatibility":     gender_compat,
+        "geo_feasibility":          geo_feasibility,
+        "med_compatibility":        med_compat,
+        "lab_availability":         float(pat_lab_score),
+        "data_completeness":        data_completeness,
+        "match_score":              match_score,
+    }
+
+
+# ---------------------------------------------------------------------------
+# Pipeline
+# ---------------------------------------------------------------------------
+
+class SecondLifePipeline:
+    """
+    Loads all data files, trains the eligibility model, and exposes
+    match_patient() for the Flask app.
+    """
+
+    def __init__(self):
+        self.model         = None
+        self.scaler        = None
+        self.model_metrics = {}
+
+        # DataFrames
+        self.patient_conditions_df = None
+        self.patient_details_df    = None
+        self.trial_studies_df      = None
+        self.trial_elig_df         = None
+        self.trial_summaries_df    = None
+        self.trial_facilities_df   = None
+
+        # Lookup structures built in load()
+        self.trial_profiles    = {}   # trial_id → profile dict
+        self.cond_to_trials    = {}   # condition → [trial_ids]
+        self.overlapping_conds = set()
+        self.patient_profiles  = {}   # patient_id → profile dict
+
+        # New real-data lookups
+        self.cond_rarity_map        = {}   # condition → rarity score
+        self.trial_us_states        = {}   # trial_id → set of US state full names
+        self.trial_facility_tokens  = {}   # trial_id → list of frozensets of facility name tokens
+        self.trial_drug_keywords    = {}   # trial_id → set of drug keywords
+        self.patient_med_keywords   = {}   # patient_id → set of med keywords
+        self.patient_lab_score      = {}   # patient_id → lab coverage (0–1)
+        self.patient_address_state  = {}   # patient_id → full US state name
+
+        self.stats = {}
+
+    # ------------------------------------------------------------------
+    # Load
+    # ------------------------------------------------------------------
+
+    def load(self):
+        print("[pipeline] Loading patient conditions …")
+        pc = pd.read_csv(PATIENT_CONDITIONS_PATH)
+        pc["condition_lower"] = pc["Condition_Name"].str.lower().str.strip()
+        self.patient_conditions_df = pc
+
+        print("[pipeline] Loading patient demographics …")
+        pd_df = pd.read_csv(PATIENT_DETAILS_PATH)
+        pd_df["Birth_Date"] = pd.to_datetime(pd_df["Birth_Date"], format="%d-%m-%Y", errors="coerce")
+        ref = datetime(2024, 1, 1)
+        pd_df["Patient_Age"] = ((ref - pd_df["Birth_Date"]).dt.days / 365.25).round()
+        pd_df["Gender"] = pd_df["Gender"].str.upper().str.strip()
+        self.patient_details_df = pd_df
+
+        if "Address" in pd_df.columns:
+            addr_series = pd_df.set_index("Patient_ID")["Address"].dropna()
+            state_series = addr_series.apply(lambda a: _extract_state(str(a)))
+            self.patient_address_state = {pid: st for pid, st in state_series.items() if st}
+            print(f"  Patients with state data: {len(self.patient_address_state):,}")
+
+        print("[pipeline] Loading patient medications …")
+        meds = pd.read_csv(PATIENT_MEDICATIONS_PATH,
+                           usecols=["Patient_ID", "Medication_Name", "Medication_End_Date"])
+        # Keep only active medications (no end date)
+        active_meds = meds[meds["Medication_End_Date"].isna()].copy()
+        active_meds["kw"] = active_meds["Medication_Name"].apply(_drug_keyword)
+        active_meds = active_meds[active_meds["kw"] != ""]
+        self.patient_med_keywords = (
+            active_meds.groupby("Patient_ID")["kw"].apply(set).to_dict()
+        )
+        # Fall back: use all meds if no active ones
+        all_meds_kw = meds.copy()
+        all_meds_kw["kw"] = all_meds_kw["Medication_Name"].apply(_drug_keyword)
+        all_meds_kw = all_meds_kw[all_meds_kw["kw"] != ""]
+        all_med_map = all_meds_kw.groupby("Patient_ID")["kw"].apply(set).to_dict()
+        for pid, kws in all_med_map.items():
+            if pid not in self.patient_med_keywords:
+                self.patient_med_keywords[pid] = kws
+        print(f"  Patients with medication data: {len(self.patient_med_keywords):,}")
+
+        print("[pipeline] Loading patient observations/labs …")
+        obs = pd.read_csv(PATIENT_OBSERVATIONS_PATH,
+                          usecols=["Patient_ID", "Observation_Name"])
+        # Unique lab types per patient, normalised by 20 (typical max)
+        lab_counts = obs.groupby("Patient_ID")["Observation_Name"].nunique()
+        self.patient_lab_score = (lab_counts / 20.0).clip(0, 1).to_dict()
+        print(f"  Patients with observation data: {len(self.patient_lab_score):,}")
+
+        print("[pipeline] Loading trial conditions …")
+        tc = pd.read_csv(TRIAL_CONDITIONS_PATH)
+        tc["condition_lower"] = tc["Condition_Name_Lower"].str.lower().str.strip()
+
+        print("[pipeline] Loading trial eligibilities …")
+        elig = pd.read_csv(TRIAL_ELIGIBILITY_PATH)
+        elig.columns = elig.columns.str.strip()
+        if "Gender" in elig.columns:
+            elig.rename(columns={"Gender": "Sex"}, inplace=True)
+        elig["Min_Age"] = elig["Minimum_Age"].apply(_parse_age).fillna(0)
+        elig["Max_Age"] = elig["Maximum_Age"].apply(_parse_age).fillna(120)
+        elig["Sex"]     = elig["Sex"].fillna("ALL").str.upper().str.strip()
+        self.trial_elig_df = elig
+
+        print("[pipeline] Loading trial studies …")
+        studies = pd.read_csv(TRIAL_STUDIES_PATH)
+        self.trial_studies_df = studies
+        print(f"  Total trials: {studies['Trial_ID'].nunique():,}")
+
+        print("[pipeline] Loading trial facilities …")
+        try:
+            fac = pd.read_csv(TRIAL_FACILITIES_PATH, encoding="utf-8", on_bad_lines="skip")
+            self.trial_facilities_df = fac
+        except Exception:
+            self.trial_facilities_df = pd.DataFrame(
+                columns=["Trial_ID", "Facility_City", "Facility_State", "Facility_Country"])
+
+        print("[pipeline] Loading trial summaries …")
+        try:
+            summaries = pd.read_csv(TRIAL_SUMMARIES_PATH, encoding="utf-8", on_bad_lines="skip")
+            self.trial_summaries_df = summaries
+        except Exception:
+            self.trial_summaries_df = pd.DataFrame(columns=["Trial_ID", "Brief_Summary"])
+
+        # ---- NEW: Trial interventions (for med_compatibility) ----
+        print("[pipeline] Loading trial interventions (drug keywords) …")
+        interv = pd.read_csv(TRIAL_INTERVENTIONS_PATH,
+                             usecols=["Trial_ID", "Intervention_Type", "Intervention_Name"])
+        drug_interv = interv[interv["Intervention_Type"] == "DRUG"].copy()
+        drug_interv["kw"] = drug_interv["Intervention_Name"].apply(_drug_keyword)
+        drug_interv = drug_interv[drug_interv["kw"] != ""]
+        self.trial_drug_keywords = (
+            drug_interv.groupby("Trial_ID")["kw"].apply(set).to_dict()
+        )
+        print(f"  Trials with drug intervention data: {len(self.trial_drug_keywords):,}")
+
+        # ---- State-level geo + facility-name index ----
+        print("[pipeline] Building trial US state index and facility-name index …")
+        fac = self.trial_facilities_df
+        if (fac is not None and not fac.empty
+                and "Facility_Country" in fac.columns
+                and "Facility_State" in fac.columns):
+            fac_us = fac[fac["Facility_Country"].str.strip().str.lower()
+                         .isin(["united states", "usa", "us"])].copy()
+            fac_us["state_clean"] = fac_us["Facility_State"].str.strip()
+            fac_us = fac_us[
+                fac_us["state_clean"].notna() &
+                (fac_us["state_clean"] != "") &
+                (fac_us["state_clean"].str.lower() != "nan")
+            ]
+            self.trial_us_states = (
+                fac_us.groupby("Trial_ID")["state_clean"].apply(set).to_dict()
+            )
+            # Facility-name token sets: trial_id → list of frozensets of significant words
+            if "Facility_Name" in fac_us.columns:
+                def _tok(name):
+                    _STOP = {"the","of","and","at","for","in","a","an","is","by",
+                             "hospital","medical","center","centre","clinic","university",
+                             "health","care","healthcare","system","institute","foundation",
+                             "research","general","regional","national","community",
+                             "services","department","division","college","school"}
+                    words = re.findall(r"[a-z]+", str(name).lower())
+                    return frozenset(w for w in words if w not in _STOP and len(w) >= 3)
+                fac_us["name_tokens"] = fac_us["Facility_Name"].apply(_tok)
+                self.trial_facility_tokens = (
+                    fac_us.groupby("Trial_ID")["name_tokens"].apply(list).to_dict()
+                )
+            else:
+                self.trial_facility_tokens = {}
+        else:
+            self.trial_us_states = {}
+            self.trial_facility_tokens = {}
+        print(f"  Trials with US facility state data: {len(self.trial_us_states):,}")
+        print(f"  Trials with facility name tokens:   {len(self.trial_facility_tokens):,}")
+
+        # ---- NEW: Trial keywords (extend condition matching bridge) ----
+        print("[pipeline] Loading trial keywords …")
+        kw_df = pd.read_csv(TRIAL_KEYWORDS_PATH,
+                            usecols=["Trial_ID", "Keyword_Name_Lower"])
+        kw_df["kw_lower"] = kw_df["Keyword_Name_Lower"].str.lower().str.strip()
+        self.trial_keyword_map = kw_df.groupby("Trial_ID")["kw_lower"].apply(set).to_dict()
+
+        # Find overlapping conditions
+        patient_cond_set = set(pc["condition_lower"].dropna().unique())
+        trial_cond_set   = set(tc["condition_lower"].dropna().unique())
+        self.overlapping_conds = patient_cond_set & trial_cond_set
+        print(f"  Overlapping conditions (exact): {len(self.overlapping_conds)}")
+
+        # ---- NEW: Condition rarity scores ----
+        print("[pipeline] Computing condition rarity scores …")
+        cond_trial_counts = tc[tc["condition_lower"].isin(self.overlapping_conds)] \
+                              .groupby("condition_lower")["Trial_ID"].nunique().to_dict()
+        for cond in self.overlapping_conds:
+            n = cond_trial_counts.get(cond, 1)
+            self.cond_rarity_map[cond] = 1.0 / math.log2(n + 2)
+        # Normalise to 0–1
+        max_r = max(self.cond_rarity_map.values()) if self.cond_rarity_map else 1.0
+        self.cond_rarity_map = {c: v / max_r for c, v in self.cond_rarity_map.items()}
+
+        # Build trial profiles
+        print("[pipeline] Building trial profiles …")
+        tc_filtered = tc[tc["condition_lower"].isin(self.overlapping_conds)]
+        trial_cond_map = tc_filtered.groupby("Trial_ID")["condition_lower"].apply(set).to_dict()
+
+        elig_idx   = elig.set_index("Trial_ID")
+        studies_idx= studies.set_index("Trial_ID")
+
+        active_statuses = {
+            "RECRUITING", "NOT_YET_RECRUITING", "ENROLLING_BY_INVITATION",
+            "ACTIVE_NOT_RECRUITING", "UNKNOWN",
+        }
+
+        for trial_id, conds in trial_cond_map.items():
+            e = elig_idx.loc[trial_id]   if trial_id in elig_idx.index   else None
+            s = studies_idx.loc[trial_id] if trial_id in studies_idx.index else None
+
+            # Handle cases where index lookup returns DataFrame (duplicate IDs)
+            if isinstance(e, pd.DataFrame): e = e.iloc[0]
+            if isinstance(s, pd.DataFrame): s = s.iloc[0]
+
+            min_age  = float(e["Min_Age"])         if e is not None else 0.0
+            max_age  = float(e["Max_Age"])         if e is not None else 120.0
+            sex      = str(e["Sex"])               if e is not None else "ALL"
+            criteria = str(e.get("Eligibility_Criteria", "")) if e is not None else ""
+
+            status   = str(s["Overall_Status"])    if s is not None else "UNKNOWN"
+            title    = str(s["Brief_Title"])       if s is not None else trial_id
+            phase    = str(s["Phase"])             if s is not None else "N/A"
+            start    = str(s.get("Start_Date", "")) if s is not None else ""
+            enroll   = s.get("Enrollment", np.nan) if s is not None else np.nan
+
+            self.trial_profiles[trial_id] = {
+                "conditions":  conds,
+                "min_age":     min_age,
+                "max_age":     max_age,
+                "sex":         sex,
+                "status":      status,
+                "is_active":   status in active_statuses,
+                "title":       title[:200],
+                "phase":       phase,
+                "start_date":  start,
+                "enrollment":  int(enroll) if pd.notna(enroll) else None,
+                "criteria":    criteria[:500],
+            }
+
+        # Reverse index: condition → [trial_ids]
+        for trial_id, prof in self.trial_profiles.items():
+            for cond in prof["conditions"]:
+                self.cond_to_trials.setdefault(cond, []).append(trial_id)
+
+        # Extend cond_to_trials with keyword matches
+        for trial_id, kws in self.trial_keyword_map.items():
+            for kw in kws:
+                if kw in self.overlapping_conds and trial_id in self.trial_profiles:
+                    self.cond_to_trials.setdefault(kw, [])
+                    if trial_id not in self.cond_to_trials[kw]:
+                        self.cond_to_trials[kw].append(trial_id)
+
+        print(f"  Trial profiles built: {len(self.trial_profiles):,}")
+
+        # Build patient profiles
+        print("[pipeline] Building patient profiles …")
+        pat_filtered = pc[pc["condition_lower"].isin(self.overlapping_conds)]
+        pat_agg = pat_filtered.groupby("Patient_ID").agg(
+            patient_conds  = ("condition_lower", set),
+            active_conds   = ("Condition_End_Date", lambda x: x.isna().sum()),
+            total_conds    = ("condition_lower", "count"),
+        ).reset_index()
+
+        details_idx = pd_df.set_index("Patient_ID")[["Patient_Age", "Gender"]]
+        pat_merged  = pat_agg.merge(details_idx, left_on="Patient_ID",
+                                    right_index=True, how="inner")
+        self.patient_profiles = pat_merged.set_index("Patient_ID").to_dict("index")
+        print(f"  Patient profiles built: {len(self.patient_profiles):,}")
+
+        self.stats = {
+            "total_patients":                pd_df["Patient_ID"].nunique(),
+            "total_trials":                  studies["Trial_ID"].nunique(),
+            "recruiting_trials":             int((studies["Overall_Status"] == "RECRUITING").sum()),
+            "overlapping_conditions":        len(self.overlapping_conds),
+            "matched_trials":                len(self.trial_profiles),
+            "patient_profiles_with_matches": len(self.patient_profiles),
+            "patients_with_medication_data": len(self.patient_med_keywords),
+            "patients_with_lab_data":        len(self.patient_lab_score),
+            "trials_with_geo_data":          len(self.trial_us_states),
+            "trials_with_drug_data":         len(self.trial_drug_keywords),
+            "_conditions":                   sorted(self.overlapping_conds),
+        }
+        return self
+
+    # ------------------------------------------------------------------
+    # Train
+    # ------------------------------------------------------------------
+
+    def train(self, n_patients: int = 3000, n_trials_per_patient: int = 30,
+              random_state: int = 42):
+        if MODEL_CACHE.exists():
+            print("[pipeline] Loading cached model …")
+            with open(MODEL_CACHE, "rb") as f:
+                cached = pickle.load(f)
+            self.model          = cached["model"]
+            self.scaler         = cached["scaler"]
+            self.model_metrics  = cached["metrics"]
+            self.overlapping_conds = cached.get("overlapping_conds", self.overlapping_conds)
+            print(f"  Cached model loaded — AUC-ROC = {self.model_metrics.get('auroc', 'N/A'):.4f}")
+            return self
+
+        print(f"[pipeline] Generating training pairs (n_patients={n_patients}) …")
+        rng = np.random.default_rng(random_state)
+
+        all_pids = list(self.patient_profiles.keys())
+        sampled_pids = rng.choice(all_pids, size=min(n_patients, len(all_pids)), replace=False)
+
+        active_tids = [tid for tid, p in self.trial_profiles.items() if p["is_active"]]
+        if not active_tids:
+            active_tids = list(self.trial_profiles.keys())
+
+        rows = []
+        for pid in sampled_pids:
+            prof       = self.patient_profiles[pid]
+            pat_conds  = prof["patient_conds"]
+            pat_age    = float(prof["Patient_Age"]) if pd.notna(prof["Patient_Age"]) else 50.0
+            pat_gender = str(prof["Gender"])        if pd.notna(prof["Gender"])       else "M"
+            total_c    = int(prof["total_conds"])
+            active_c   = int(prof["active_conds"])
+
+            pat_med_kws  = self.patient_med_keywords.get(pid, set())
+            pat_lab      = self.patient_lab_score.get(pid, 0.3)
+            pat_state    = self.patient_address_state.get(pid, "")
+
+            # Positive candidates: share at least one condition
+            candidate_tids = set()
+            for c in pat_conds:
+                candidate_tids.update(self.cond_to_trials.get(c, []))
+
+            if not candidate_tids:
+                continue
+
+            # Negative candidates: random trials with no condition overlap
+            neg_tids = set(rng.choice(active_tids,
+                                       size=min(10, len(active_tids)),
+                                       replace=False).tolist())
+
+            candidate_list = list(candidate_tids)
+            rng.shuffle(candidate_list)
+            candidate_list = candidate_list[:n_trials_per_patient]
+            candidate_list += [t for t in neg_tids if t not in candidate_tids]
+
+            for tid in candidate_list:
+                if tid not in self.trial_profiles:
+                    continue
+                tp = self.trial_profiles[tid]
+
+                feats = _compute_features(
+                    pat_conds, pat_age, pat_gender,
+                    tp["conditions"], tp["min_age"], tp["max_age"], tp["sex"],
+                    total_c, active_c, len(tp["conditions"]),
+                    self.cond_rarity_map,
+                    self.trial_us_states.get(tid, set()),
+                    pat_med_kws,
+                    self.trial_drug_keywords.get(tid, set()),
+                    pat_lab,
+                    patient_state_full=pat_state,
+                )
+                feats["Patient_ID"] = pid
+                feats["Trial_ID"]   = tid
+                rows.append(feats)
+
+        df = pd.DataFrame(rows)
+        print(f"  Training pairs generated: {len(df):,}")
+
+        # Rule-based label — uses all 6 main signals, 15% noise for realism
+        import hashlib
+        def _label(row):
+            score = (
+                0.30 * float(row["age_compatibility"]    > 0.6) +
+                0.15 * float(row["gender_compatibility"] > 0.5) +
+                0.25 * float(row["jaccard_similarity"]   > 0.05) +
+                0.10 * float(row["geo_feasibility"]) +
+                0.10 * float(row["med_compatibility"]) +
+                0.10 * float(row["lab_availability"])
+            )
+            base = int(score >= 0.5)
+            h = int(hashlib.md5(
+                    f"{row['Patient_ID']}_{row['Trial_ID']}".encode()
+                ).hexdigest()[:8], 16)
+            if (h % 1000) / 1000 < 0.15:
+                base = 1 - base
+            return base
+
+        df["label"] = df.apply(_label, axis=1)
+
+        X      = df[FEATURE_COLS].fillna(0)
+        y      = df["label"].values
+        groups = df["Patient_ID"].values
+
+        gss = GroupShuffleSplit(n_splits=1, test_size=0.2, random_state=random_state)
+        train_idx, test_idx = next(gss.split(X, y, groups))
+
+        X_train, X_test = X.iloc[train_idx], X.iloc[test_idx]
+        y_train, y_test = y[train_idx], y[test_idx]
+        g_train         = groups[train_idx]
+
+        print(f"  Train: {len(X_train):,} | Test: {len(X_test):,} | Positive rate: {y_train.mean()*100:.1f}%")
+
+        scaler     = StandardScaler()
+        X_train_sc = scaler.fit_transform(X_train)
+
+        print("[pipeline] Training Random Forest (n_estimators=200, max_depth=12) …")
+        rf = RandomForestClassifier(
+            n_estimators=200, max_depth=12, min_samples_split=10,
+            class_weight="balanced", random_state=random_state, n_jobs=-1,
+        )
+        rf.fit(X_train, y_train)
+
+        # Calibrate probabilities
+        cal = CalibratedClassifierCV(rf, cv=3, method="isotonic")
+        cal.fit(X_train, y_train)
+
+        y_pred = cal.predict(X_test)
+        y_prob = cal.predict_proba(X_test)[:, 1]
+
+        metrics = {
+            "accuracy":      accuracy_score(y_test, y_pred),
+            "precision":     precision_score(y_test, y_pred, zero_division=0),
+            "recall":        recall_score(y_test, y_pred, zero_division=0),
+            "f1":            f1_score(y_test, y_pred, zero_division=0),
+            "auroc":         roc_auc_score(y_test, y_prob),
+            "brier":         brier_score_loss(y_test, y_prob),
+            "avg_precision": average_precision_score(y_test, y_prob),
+            "train_size":    len(X_train),
+            "test_size":     len(X_test),
+            "positive_rate": float(y_train.mean()),
+        }
+
+        cv = GroupKFold(n_splits=5)
+        cv_scores = cross_val_score(rf, X_train, y_train, cv=cv,
+                                     groups=g_train, scoring="roc_auc")
+        metrics["cv_auroc_mean"] = float(cv_scores.mean())
+        metrics["cv_auroc_std"]  = float(cv_scores.std())
+        metrics["feature_importance"] = dict(zip(FEATURE_COLS, rf.feature_importances_))
+
+        print(f"  Accuracy : {metrics['accuracy']*100:.1f}%")
+        print(f"  AUC-ROC  : {metrics['auroc']:.4f}")
+        print(f"  CV AUC   : {metrics['cv_auroc_mean']:.4f} ± {metrics['cv_auroc_std']:.4f}")
+
+        self.model         = cal
+        self.scaler        = scaler
+        self.model_metrics = metrics
+
+        with open(MODEL_CACHE, "wb") as f:
+            pickle.dump({
+                "model":             cal,
+                "scaler":            scaler,
+                "metrics":           metrics,
+                "overlapping_conds": self.overlapping_conds,
+            }, f)
+        print("[pipeline] Model saved to cache.")
+        return self
+
+    # ------------------------------------------------------------------
+    # Match
+    # ------------------------------------------------------------------
+
+    def match_patient(self, conditions: list, age: float, gender: str,
+                      top_k: int = 20, active_only: bool = True,
+                      patient_id: str = None, address: str = "") -> list:
+        if self.model is None:
+            raise RuntimeError("Model not trained. Call .train() first.")
+
+        input_conds   = {c.lower().strip() for c in conditions}
+        matched_conds = input_conds & self.overlapping_conds
+        if not matched_conds:
+            return []
+
+        # Candidate trials
+        candidate_tids = set()
+        for c in matched_conds:
+            candidate_tids.update(self.cond_to_trials.get(c, []))
+        if not candidate_tids:
+            return []
+
+        total_c  = len(input_conds)
+        active_c = int(total_c * 0.7)
+        gender   = str(gender).upper().strip()
+
+        pat_med_kws = self.patient_med_keywords.get(patient_id, set()) if patient_id else set()
+        pat_lab     = self.patient_lab_score.get(patient_id, 0.3)     if patient_id else 0.3
+
+        # State-level geo: prefer provided address, fall back to Synthea lookup
+        patient_state_full = _extract_state(address) if address else ""
+        if not patient_state_full and patient_id:
+            patient_state_full = self.patient_address_state.get(patient_id, "")
+
+        rows = []
+        for tid in candidate_tids:
+            if tid not in self.trial_profiles:
+                continue
+            tp = self.trial_profiles[tid]
+            if active_only and not tp["is_active"]:
+                continue
+
+            feats = _compute_features(
+                matched_conds, age, gender,
+                tp["conditions"], tp["min_age"], tp["max_age"], tp["sex"],
+                total_c, active_c, len(tp["conditions"]),
+                self.cond_rarity_map,
+                self.trial_us_states.get(tid, set()),
+                pat_med_kws,
+                self.trial_drug_keywords.get(tid, set()),
+                pat_lab,
+                patient_state_full=patient_state_full,
+            )
+            feats["Trial_ID"] = tid
+            rows.append(feats)
+
+        if not rows:
+            return []
+
+        df = pd.DataFrame(rows)
+        X  = df[FEATURE_COLS].fillna(0)
+
+        probs = self.model.predict_proba(X)[:, 1]
+        df["eligibility_probability"] = probs
+        df["combined_score"] = 0.6 * probs + 0.4 * (df["match_score"] / 100.0)
+
+        df_sorted = df.sort_values("combined_score", ascending=False).head(top_k)
+
+        results = []
+        for _, row in df_sorted.iterrows():
+            tid  = row["Trial_ID"]
+            tp   = self.trial_profiles[tid]
+            overlap_conds = sorted(matched_conds & tp["conditions"])
+
+            summary = ""
+            if self.trial_summaries_df is not None and not self.trial_summaries_df.empty:
+                s_rows = self.trial_summaries_df[self.trial_summaries_df["Trial_ID"] == tid]
+                if not s_rows.empty:
+                    col = "Brief_Summary" if "Brief_Summary" in s_rows.columns else s_rows.columns[-1]
+                    summary = str(s_rows.iloc[0][col])[:400]
+
+            location = ""
+            facility_name = ""
+            n_sites  = 0
+            if self.trial_facilities_df is not None and not self.trial_facilities_df.empty:
+                f_rows = self.trial_facilities_df[self.trial_facilities_df["Trial_ID"] == tid]
+                if not f_rows.empty:
+                    # Prefer US sites first, fall back to first row
+                    us_rows = f_rows[f_rows.get("Facility_Country", pd.Series(dtype=str))
+                                     .str.strip().str.lower()
+                                     .isin(["united states", "usa", "us"])] \
+                              if "Facility_Country" in f_rows.columns else pd.DataFrame()
+                    r = us_rows.iloc[0] if not us_rows.empty else f_rows.iloc[0]
+                    parts = [str(r.get(c, "")) for c in
+                             ["Facility_City", "Facility_State", "Facility_Country"]
+                             if str(r.get(c, "")).strip() not in ("", "nan")]
+                    location = ", ".join(parts)
+                    if "Facility_Name" in r.index:
+                        fn = str(r.get("Facility_Name", "")).strip()
+                        facility_name = fn if fn not in ("", "nan") else ""
+                    n_sites  = len(f_rows)
+
+            results.append({
+                "trial_id":                tid,
+                "title":                   tp["title"],
+                "phase":                   tp["phase"],
+                "status":                  tp["status"],
+                "min_age":                 int(tp["min_age"]),
+                "max_age":                 int(tp["max_age"]),
+                "sex":                     tp["sex"],
+                "enrollment":              tp["enrollment"],
+                "start_date":              tp["start_date"],
+                "eligibility_probability": round(float(row["eligibility_probability"]) * 100, 1),
+                "match_score":             round(float(row["match_score"]), 1),
+                "combined_score":          round(float(row["combined_score"]) * 100, 1),
+                "jaccard_similarity":      round(float(row["jaccard_similarity"]), 3),
+                "age_compatibility":       round(float(row["age_compatibility"]) * 100, 1),
+                "gender_compatibility":    round(float(row["gender_compatibility"]) * 100, 1),
+                "geo_feasibility":         round(float(row["geo_feasibility"]) * 100, 1),
+                "med_compatibility":       round(float(row["med_compatibility"]) * 100, 1),
+                "condition_rarity_score":  round(float(row["condition_rarity_score"]), 3),
+                "overlap_conditions":      overlap_conds,
+                "trial_conditions":        sorted(tp["conditions"]),
+                "criteria":                tp["criteria"],
+                "summary":                 summary,
+                "location":                location,
+                "facility_name":           facility_name,
+                "n_sites":                 n_sites,
+            })
+
+        return results
+
+    # ------------------------------------------------------------------
+    # Hospital trial dashboard
+    # ------------------------------------------------------------------
+
+    def trials_for_hospital(self, hospital_name: str, location: str,
+                             research_conditions: list, top_k: int = 20) -> list:
+        """
+        Return trials relevant to this hospital using 3-tier logic (reversed from
+        hospitals-for-trial): Tier 1 = Jaccard name match, Tier 2 = same state,
+        Tier 3 = research-condition overlap.
+        """
+        h_tokens = _name_tokens(hospital_name)
+
+        m = re.search(r",\s*([A-Z]{2})\s*$", str(location).strip())
+        h_state = STATE_ABBREV.get(m.group(1), "") if m else ""
+        rc_set  = {r.lower().strip() for r in (research_conditions or []) if r}
+
+        collected: list[tuple[str, int]] = []
+        seen: set = set()
+
+        # Tier 1: facility-name Jaccard ≥ 0.25
+        for trial_id, fac_list in self.trial_facility_tokens.items():
+            if trial_id not in self.trial_profiles:
+                continue
+            if not self.trial_profiles[trial_id].get("is_active", False):
+                continue
+            best = 0.0
+            for ft in fac_list:
+                if h_tokens and ft:
+                    inter = len(h_tokens & ft)
+                    union = len(h_tokens | ft)
+                    s = inter / union if union else 0.0
+                    if s > best:
+                        best = s
+            if best >= 0.25:
+                collected.append((trial_id, 1))
+                seen.add(trial_id)
+
+        # Tier 2: same state — cap at 2×top_k additional to avoid runaway
+        if h_state and len(collected) < top_k:
+            t2_cap = top_k * 2
+            t2_added = 0
+            for trial_id, states in self.trial_us_states.items():
+                if t2_added >= t2_cap:
+                    break
+                if trial_id in seen or trial_id not in self.trial_profiles:
+                    continue
+                if not self.trial_profiles[trial_id].get("is_active", False):
+                    continue
+                if h_state in states:
+                    collected.append((trial_id, 2))
+                    seen.add(trial_id)
+                    t2_added += 1
+
+        # Tier 3: research-condition overlap — fill up to top_k
+        if rc_set and len(collected) < top_k:
+            t3_cap = top_k * 2
+            t3_added = 0
+            for cond in rc_set:
+                for trial_id in self.cond_to_trials.get(cond, []):
+                    if t3_added >= t3_cap:
+                        break
+                    if trial_id in seen or trial_id not in self.trial_profiles:
+                        continue
+                    if not self.trial_profiles[trial_id].get("is_active", False):
+                        continue
+                    collected.append((trial_id, 3))
+                    seen.add(trial_id)
+                    t3_added += 1
+
+        collected.sort(key=lambda x: x[1])
+        collected = collected[:top_k]
+
+        _REASON = {1: "verified trial site", 2: "trial in your state",
+                   3: "matches your research conditions"}
+
+        output = []
+        for trial_id, tier in collected:
+            tp = self.trial_profiles[trial_id]
+
+            summary = ""
+            if self.trial_summaries_df is not None and not self.trial_summaries_df.empty:
+                s_rows = self.trial_summaries_df[self.trial_summaries_df["Trial_ID"] == trial_id]
+                if not s_rows.empty:
+                    col = "Brief_Summary" if "Brief_Summary" in s_rows.columns else s_rows.columns[-1]
+                    summary = str(s_rows.iloc[0][col])[:300]
+
+            location_str = facility_name = ""
+            n_sites = 0
+            if self.trial_facilities_df is not None and not self.trial_facilities_df.empty:
+                f_rows = self.trial_facilities_df[self.trial_facilities_df["Trial_ID"] == trial_id]
+                if not f_rows.empty:
+                    n_sites = len(f_rows)
+                    us_rows = f_rows[f_rows["Facility_Country"].str.strip().str.lower()
+                                     .isin(["united states", "usa", "us"])] \
+                              if "Facility_Country" in f_rows.columns else pd.DataFrame()
+                    r = us_rows.iloc[0] if not us_rows.empty else f_rows.iloc[0]
+                    parts = [str(r.get(c, "")) for c in
+                             ["Facility_City", "Facility_State", "Facility_Country"]
+                             if str(r.get(c, "")).strip() not in ("", "nan")]
+                    location_str = ", ".join(parts)
+                    if "Facility_Name" in r.index:
+                        fn = str(r.get("Facility_Name", "")).strip()
+                        facility_name = fn if fn not in ("", "nan") else ""
+
+            output.append({
+                "trial_id":     trial_id,
+                "title":        tp["title"],
+                "phase":        tp["phase"],
+                "status":       tp["status"],
+                "min_age":      int(tp["min_age"]),
+                "max_age":      int(tp["max_age"]),
+                "sex":          tp["sex"],
+                "enrollment":   tp["enrollment"],
+                "conditions":   sorted(tp["conditions"]),
+                "summary":      summary,
+                "location":     location_str,
+                "facility_name": facility_name,
+                "n_sites":      n_sites,
+                "match_tier":   tier,
+                "match_reason": _REASON[tier],
+            })
+
+        return output
+
+    # ------------------------------------------------------------------
+    # Patient lookup
+    # ------------------------------------------------------------------
+
+    def get_patient(self, patient_id: str):
+        if patient_id not in self.patient_profiles:
+            return None
+        prof = self.patient_profiles[patient_id]
+        det  = self.patient_details_df[self.patient_details_df["Patient_ID"] == patient_id]
+        if det.empty:
+            return None
+        row = det.iloc[0]
+
+        all_conds = sorted(
+            self.patient_conditions_df[
+                self.patient_conditions_df["Patient_ID"] == patient_id
+            ]["Condition_Name"].dropna().unique().tolist()
+        )
+
+        return {
+            "patient_id":            patient_id,
+            "age":                   int(prof["Patient_Age"]) if pd.notna(prof["Patient_Age"]) else None,
+            "gender":                str(prof["Gender"]),
+            "first_name":            str(row.get("First_Name", "")),
+            "last_name":             str(row.get("Last_Name", "")),
+            "race":                  str(row.get("Race", "")),
+            "ethnicity":             str(row.get("Ethnicity", "")),
+            "address":               str(row.get("Address", "")),
+            "conditions":            all_conds,
+            "matching_conditions":   sorted(prof["patient_conds"]),
+            "medications":           sorted(self.patient_med_keywords.get(patient_id, set())),
+            "has_lab_data":          patient_id in self.patient_lab_score,
+        }
+
+    def search_patients(self, query: str, limit: int = 20) -> list:
+        query_l = query.lower().strip()
+        results = []
+        for pid, prof in self.patient_profiles.items():
+            if query_l in pid.lower():
+                det = self.patient_details_df[self.patient_details_df["Patient_ID"] == pid]
+                if not det.empty:
+                    r = det.iloc[0]
+                    results.append({
+                        "patient_id":   pid,
+                        "age":          int(prof["Patient_Age"]) if pd.notna(prof["Patient_Age"]) else None,
+                        "gender":       str(prof["Gender"]),
+                        "first_name":   str(r.get("First_Name", "")),
+                        "last_name":    str(r.get("Last_Name", "")),
+                        "n_conditions": len(prof["patient_conds"]),
+                    })
+                if len(results) >= limit:
+                    break
+        return results
+
+    # ------------------------------------------------------------------
+    # MIMIC Validation
+    # ------------------------------------------------------------------
+
+    def _map_icd_to_overlapping(self, icd_long_titles: list) -> list:
+        matched = set()
+        for title in icd_long_titles:
+            t = title.lower().strip()
+            if t in self.overlapping_conds:
+                matched.add(t)
+                continue
+            for cond in self.overlapping_conds:
+                if cond in t or t in cond:
+                    matched.add(cond)
+                    break
+            t_words = set(t.split())
+            for cond in self.overlapping_conds:
+                c_words  = set(cond.split())
+                shorter  = min(len(t_words), len(c_words))
+                if shorter > 0 and len(t_words & c_words) / shorter >= 0.75:
+                    matched.add(cond)
+        return list(matched)
+
+    def validate_mimic(self) -> list:
+        print("[pipeline] Running MIMIC-IV validation …")
+        patients  = pd.read_csv(MIMIC_PATIENTS_PATH)
+        diagnoses = pd.read_csv(MIMIC_DIAGNOSES_PATH)
+        icd_dict  = pd.read_csv(MIMIC_ICD_DICT_PATH)
+
+        icd_map = icd_dict.set_index(["icd_code", "icd_version"])["long_title"].to_dict()
+        diagnoses["condition_name"] = diagnoses.apply(
+            lambda r: icd_map.get((r["icd_code"], r["icd_version"]),
+                                   str(r["icd_code"])).lower(), axis=1
+        )
+
+        results = []
+        for _, pat in patients.iterrows():
+            sid    = int(pat["subject_id"])
+            age    = float(pat["anchor_age"])
+            gender = "M" if pat["gender"] == "M" else "F"
+
+            raw_conds    = diagnoses[diagnoses["subject_id"] == sid]["condition_name"].dropna().unique().tolist()
+            mapped_conds = self._map_icd_to_overlapping(raw_conds)
+            all_conds    = list(set(raw_conds + mapped_conds))
+
+            matches = self.match_patient(all_conds, age, gender, top_k=5)
+
+            results.append({
+                "subject_id":         sid,
+                "age":                int(age),
+                "gender":             gender,
+                "n_conditions":       len(raw_conds),
+                "conditions":         raw_conds[:5],
+                "mapped_conditions":  mapped_conds[:5],
+                "n_matches":          len(matches),
+                "top_match":          matches[0] if matches else None,
+            })
+        return results
+
+    # ------------------------------------------------------------------
+    # Autocomplete
+    # ------------------------------------------------------------------
+
+    def condition_autocomplete(self, query: str, limit: int = 15) -> list:
+        q = query.lower().strip()
+        return sorted([c for c in self.overlapping_conds if q in c])[:limit]

requirements.txt

@@ -0,0 +1,6 @@
+Flask==3.1.3
+Werkzeug==3.1.3
+numpy==2.4.3
+pandas==3.0.1
+scikit-learn==1.8.0
+huggingface_hub==1.11.0

static/.gitkeep

@@ -0,0 +1 @@
+

templates/hospital.html

@@ -0,0 +1,665 @@
+<!DOCTYPE html>
+<html lang="en" data-bs-theme="dark">
+<head>
+<meta charset="UTF-8">
+<meta name="viewport" content="width=device-width, initial-scale=1.0">
+<title>Hospital Portal — Second Life</title>
+<link href="https://cdn.jsdelivr.net/npm/bootstrap@5.3.2/dist/css/bootstrap.min.css" rel="stylesheet">
+<style>
+body{background:#0d1117;}
+.patient-card{background:#161b22;border:1px solid #30363d;border-radius:10px;padding:1.25rem;height:100%;}
+.form-control,.form-select{background:#0d1117 !important;border-color:#30363d;color:#e6edf3 !important;}
+.form-control:focus{border-color:#58a6ff !important;box-shadow:0 0 0 .2rem rgba(88,166,255,.2) !important;}
+.autocomplete-list{position:absolute;z-index:1000;width:100%;background:#161b22;border:1px solid #30363d;
+  border-radius:0 0 6px 6px;max-height:190px;overflow-y:auto;display:none;list-style:none;padding:0;margin:0;}
+.autocomplete-list li{padding:.3rem .6rem;cursor:pointer;font-size:.85rem;color:#c9d1d9;}
+.autocomplete-list li:hover{background:#21262d;}
+.st-pending{color:#d29922;}.st-accepted{color:#2ea043;}.st-rejected{color:#f85149;}.st-completed{color:#58a6ff;}
+</style>
+</head>
+<body>
+<nav class="navbar border-bottom border-secondary px-3 py-2" style="background:#161b22">
+  <span class="navbar-brand fw-bold" style="color:#58a6ff">⬡ Second Life</span>
+  <div class="d-flex align-items-center gap-3">
+    <span class="text-light small">{{ hospital.hospital_name }}</span>
+    <span class="badge bg-success">Hospital</span>
+    <button class="btn btn-sm btn-outline-secondary" id="logoutBtn">Logout</button>
+  </div>
+</nav>
+
+<div class="container-xl py-3">
+  <ul class="nav nav-tabs mb-4" id="mainTabs">
+    <li class="nav-item"><button class="nav-link active" data-tab="patients">Available Patients</button></li>
+    <li class="nav-item"><button class="nav-link" data-tab="search">Search by Condition</button></li>
+    <li class="nav-item"><button class="nav-link" data-tab="trials">My Trials</button></li>
+    <li class="nav-item"><button class="nav-link" data-tab="inbox">Inbox <span id="inboxBadge" class="badge bg-danger ms-1" style="display:none">0</span></button></li>
+    <li class="nav-item"><button class="nav-link" data-tab="connections">My Connections</button></li>
+    <li class="nav-item"><button class="nav-link" data-tab="profile">My Profile</button></li>
+  </ul>
+
+  <!-- ===== AVAILABLE PATIENTS ===== -->
+  <div id="tab-patients">
+    <div class="d-flex justify-content-between align-items-center mb-2">
+      <div>
+        <h6 class="mb-0">Patients Open to Trials</h6>
+        <small class="text-muted">Patients who have opted in and are not yet connected to your hospital</small>
+      </div>
+      <button class="btn btn-sm btn-primary" id="refreshPatientsBtn">Refresh</button>
+    </div>
+    <div class="alert alert-secondary py-2 small mb-3">
+      Only patients with <strong>Open to Trials</strong> toggled on and no existing connection with your hospital appear here.
+    </div>
+    <div id="pLoading" class="text-center py-4" style="display:none">
+      <div class="spinner-border text-primary spinner-border-sm"></div>
+    </div>
+    <div id="pEmpty" class="text-center py-5 text-muted" style="display:none">
+      No patients are currently open to trials.
+    </div>
+    <div class="row g-3" id="pGrid"></div>
+  </div>
+
+  <!-- ===== SEARCH ===== -->
+  <div id="tab-search" style="display:none">
+    <div class="row g-3 align-items-end mb-3">
+      <div class="col-md-6">
+        <label class="form-label small fw-bold">Search by Medical Condition</label>
+        <div class="position-relative">
+          <input type="text" class="form-control" id="searchInput"
+                 placeholder="e.g. hypertension, diabetes…" autocomplete="off">
+          <ul class="autocomplete-list" id="searchAutoList"></ul>
+        </div>
+      </div>
+      <div class="col-md-2">
+        <button class="btn btn-primary w-100" id="searchBtn">Search</button>
+      </div>
+      <div class="col-md-4">
+        <div class="form-check mb-1">
+          <input class="form-check-input" type="checkbox" id="searchIncludeConnected">
+          <label class="form-check-label small" for="searchIncludeConnected">
+            Include patients already connected to your hospital
+          </label>
+        </div>
+        <small class="text-muted">Default: only unconnected patients shown</small>
+      </div>
+    </div>
+    <div id="sLoading" class="text-center py-4" style="display:none">
+      <div class="spinner-border text-primary spinner-border-sm"></div>
+    </div>
+    <div id="sEmpty" class="text-center py-5 text-muted" style="display:none">
+      No matching patients found for this condition.
+    </div>
+    <div class="row g-3" id="sGrid"></div>
+  </div>
+
+  <!-- ===== MY TRIALS ===== -->
+  <div id="tab-trials" style="display:none">
+    <div class="d-flex justify-content-between align-items-center mb-3">
+      <div>
+        <h6 class="mb-0">Trials Relevant to Your Hospital</h6>
+        <small class="text-muted">Matched by facility name, state, or your research conditions</small>
+      </div>
+      <button class="btn btn-sm btn-primary" id="refreshTrialsBtn">Refresh</button>
+    </div>
+    <div id="tLoading" class="text-center py-4" style="display:none">
+      <div class="spinner-border text-primary spinner-border-sm"></div>
+    </div>
+    <div id="tEmpty" class="text-center py-5 text-muted" style="display:none">
+      No trials found. Update your profile with your location and research conditions.
+    </div>
+    <div class="row g-2" id="tGrid"></div>
+  </div>
+
+  <!-- ===== INBOX ===== -->
+  <div id="tab-inbox" style="display:none">
+    <div class="d-flex justify-content-between align-items-center mb-3">
+      <h6 class="mb-0">Inbox</h6>
+      <button class="btn btn-sm btn-outline-secondary" onclick="loadInbox()">Refresh</button>
+    </div>
+    <div id="inboxLoading" class="text-center text-muted py-4">Loading…</div>
+    <div id="inboxEmpty" class="text-center text-muted py-4" style="display:none">No message threads yet. Accept a connection and start chatting.</div>
+    <div id="inboxList"></div>
+  </div>
+
+  <!-- ===== PROFILE ===== -->
+  <div id="tab-profile" style="display:none">
+    <div class="row justify-content-center">
+      <div class="col-lg-6">
+        <div class="patient-card">
+          <h6 class="mb-3">Hospital Profile</h6>
+          <div class="mb-3">
+            <label class="form-label small fw-bold">Hospital / Institution Name</label>
+            <input type="text" class="form-control" id="profName">
+          </div>
+          <div class="mb-3">
+            <label class="form-label small fw-bold">Location <span class="text-muted fw-normal">(City, ST — e.g. Boston, MA)</span></label>
+            <input type="text" class="form-control" id="profLoc">
+          </div>
+          <div class="mb-3">
+            <label class="form-label small fw-bold">Research Conditions</label>
+            <div class="d-flex gap-2 mb-2">
+              <input type="text" class="form-control form-control-sm" id="profCondInput"
+                     placeholder="Type a condition and press Enter" autocomplete="off">
+            </div>
+            <div id="profCondTags" class="d-flex flex-wrap gap-1 mb-1"></div>
+          </div>
+          <div id="profMsg" class="small mb-2" style="min-height:1.2rem"></div>
+          <button class="btn btn-success" id="saveProfileBtn">Save Changes</button>
+        </div>
+      </div>
+    </div>
+  </div>
+
+  <!-- ===== CONNECTIONS ===== -->
+  <div id="tab-connections" style="display:none">
+    <div class="d-flex justify-content-between align-items-center mb-3">
+      <h6 class="mb-0">Connection Requests</h6>
+      <button class="btn btn-sm btn-outline-secondary" id="refreshConnsBtn">Refresh</button>
+    </div>
+    <div id="cEmpty" class="text-center py-5 text-muted" style="display:none">
+      No connection requests yet.
+    </div>
+    <div id="cWrap" style="display:none">
+      <div class="table-responsive">
+        <table class="table table-dark table-hover table-sm align-middle">
+          <thead><tr>
+            <th>Patient</th><th>Age/Gender</th><th>Conditions</th>
+            <th>Trial</th><th>Initiated</th><th>Status</th><th>Message</th><th>Action</th><th>Chat</th>
+          </tr></thead>
+          <tbody id="cTbody"></tbody>
+        </table>
+      </div>
+    </div>
+  </div>
+</div>
+
+<!-- Connect with Patient Modal -->
+<div class="modal fade" id="connectModal" tabindex="-1">
+  <div class="modal-dialog">
+    <div class="modal-content border-secondary" style="background:#161b22">
+      <div class="modal-header border-secondary">
+        <h5 class="modal-title">Connect with Patient</h5>
+        <button type="button" class="btn-close btn-close-white" data-bs-dismiss="modal"></button>
+      </div>
+      <div class="modal-body">
+        <p class="text-muted small" id="modalPatientInfo"></p>
+        <label class="form-label small">Message <span class="text-muted">(optional)</span></label>
+        <textarea class="form-control form-control-sm" id="modalMsg" rows="3"
+                  placeholder="Introduce your trial or research program…"></textarea>
+        <div class="alert alert-info py-2 small mt-3">
+          The patient will receive your request and can choose to accept or decline.
+        </div>
+      </div>
+      <div class="modal-footer border-secondary">
+        <button class="btn btn-secondary btn-sm" data-bs-dismiss="modal">Cancel</button>
+        <button class="btn btn-success btn-sm" id="sendConnBtn">Send Request</button>
+      </div>
+    </div>
+  </div>
+</div>
+
+<!-- Messages Modal -->
+<div class="modal fade" id="msgModal" tabindex="-1">
+  <div class="modal-dialog modal-lg">
+    <div class="modal-content border-secondary" style="background:#161b22">
+      <div class="modal-header border-secondary">
+        <h5 class="modal-title" id="msgModalTitle">Messages</h5>
+        <button type="button" class="btn-close btn-close-white" data-bs-dismiss="modal"></button>
+      </div>
+      <div class="modal-body p-0">
+        <div id="msgThread" style="height:320px;overflow-y:auto;padding:1rem;background:#0d1117"></div>
+        <div class="p-3 border-top border-secondary d-flex gap-2">
+          <input type="text" class="form-control form-control-sm" id="msgInput"
+                 placeholder="Type a message…" maxlength="2000">
+          <button class="btn btn-success btn-sm px-3" id="msgSendBtn">Send</button>
+        </div>
+      </div>
+    </div>
+  </div>
+</div>
+
+<script src="https://cdn.jsdelivr.net/npm/bootstrap@5.3.2/dist/js/bootstrap.bundle.min.js"></script>
+<script>
+const HOSPITAL = {{ hospital | tojson }};
+let connectModal, msgModal, targetPatientId, currentConnId;
+let profConds = [];
+
+document.addEventListener('DOMContentLoaded', () => {
+  connectModal = new bootstrap.Modal(document.getElementById('connectModal'));
+  msgModal     = new bootstrap.Modal(document.getElementById('msgModal'));
+
+  document.querySelectorAll('#mainTabs .nav-link').forEach(b =>
+    b.addEventListener('click', () => switchTab(b.dataset.tab)));
+  document.getElementById('logoutBtn').addEventListener('click', logout);
+  document.getElementById('refreshPatientsBtn').addEventListener('click', loadPatients);
+  document.getElementById('refreshConnsBtn').addEventListener('click', loadConnections);
+  document.getElementById('refreshTrialsBtn').addEventListener('click', loadTrials);
+  document.getElementById('searchBtn').addEventListener('click', doSearch);
+  document.getElementById('sendConnBtn').addEventListener('click', submitConnect);
+  document.getElementById('saveProfileBtn').addEventListener('click', saveProfile);
+  document.getElementById('msgSendBtn').addEventListener('click', sendMsg);
+  document.getElementById('msgInput').addEventListener('keydown', e => {
+    if (e.key === 'Enter') sendMsg();
+  });
+
+  // Profile condition tag input
+  const pci = document.getElementById('profCondInput');
+  pci.addEventListener('keydown', e => {
+    if (e.key === 'Enter' || e.key === ',') {
+      e.preventDefault();
+      const val = pci.value.trim().toLowerCase();
+      if (val && !profConds.includes(val)) { profConds.push(val); renderProfTags(); }
+      pci.value = '';
+    }
+  });
+
+  // Condition autocomplete (search tab)
+  const si = document.getElementById('searchInput');
+  si.addEventListener('input', () => condAuto(si.value));
+  si.addEventListener('keydown', e => { if(e.key==='Enter') doSearch(); if(e.key==='Escape') hideAuto(); });
+  document.addEventListener('click', e => { if (!si.contains(e.target)) hideAuto(); });
+
+  // Event delegation for connections table
+  document.getElementById('cTbody').addEventListener('click', e => {
+    const btn = e.target.closest('[data-cid]');
+    if (!btn) return;
+    if (btn.dataset.st)                     updateStatus(btn.dataset.cid, btn.dataset.st);
+    else if ('chat' in btn.dataset)         openMsgModal(btn.dataset.cid, btn.dataset.chat);
+  });
+
+  loadPatients();
+});
+
+function switchTab(name) {
+  document.querySelectorAll('#mainTabs .nav-link').forEach(b =>
+    b.classList.toggle('active', b.dataset.tab === name));
+  ['patients','search','trials','inbox','connections','profile'].forEach(t =>
+    document.getElementById('tab-'+t).style.display = t === name ? '' : 'none');
+  if (name === 'connections') loadConnections();
+  if (name === 'profile')     initProfile();
+  if (name === 'trials')      loadTrials();
+  if (name === 'inbox')       loadInbox();
+}
+
+// ── PROFILE ──────────────────────────────────────────────────────────────────
+function initProfile() {
+  document.getElementById('profName').value = HOSPITAL.hospital_name || '';
+  document.getElementById('profLoc').value  = HOSPITAL.location || '';
+  profConds = Array.isArray(HOSPITAL.research_conditions) ? [...HOSPITAL.research_conditions] : [];
+  renderProfTags();
+  document.getElementById('profMsg').textContent = '';
+}
+
+function renderProfTags() {
+  const wrap = document.getElementById('profCondTags');
+  wrap.innerHTML = '';
+  profConds.forEach((c, i) => {
+    const span = document.createElement('span');
+    span.className = 'badge bg-info text-dark d-inline-flex align-items-center gap-1';
+    const t = document.createElement('span'); t.textContent = c;
+    const x = document.createElement('button');
+    x.type = 'button'; x.className = 'btn-close'; x.style.fontSize = '.45rem';
+    x.addEventListener('click', () => { profConds.splice(i, 1); renderProfTags(); });
+    span.appendChild(t); span.appendChild(x); wrap.appendChild(span);
+  });
+}
+
+async function saveProfile() {
+  const body = {
+    hospital_name:       document.getElementById('profName').value.trim(),
+    location:            document.getElementById('profLoc').value.trim(),
+    research_conditions: profConds,
+  };
+  const msg = document.getElementById('profMsg');
+  msg.textContent = 'Saving…'; msg.style.color = '#8b949e';
+  try {
+    const r = await fetch('/api/hospital/profile', {
+      method: 'POST', headers: {'Content-Type': 'application/json'},
+      body: JSON.stringify(body),
+    });
+    const d = await r.json();
+    if (r.ok) {
+      // Update the in-memory HOSPITAL object so navbar reflects change
+      HOSPITAL.hospital_name        = d.hospital_name;
+      HOSPITAL.location             = d.location;
+      HOSPITAL.research_conditions  = d.research_conditions;
+      document.querySelector('.navbar .text-light.small').textContent = d.hospital_name;
+      msg.textContent = 'Saved.'; msg.style.color = '#2ea043';
+    } else {
+      msg.textContent = d.error || 'Save failed'; msg.style.color = '#f85149';
+    }
+  } catch(e) { msg.textContent = 'Network error'; msg.style.color = '#f85149'; }
+}
+
+// ── PATIENTS ─────────────────────────────────────────────────────────────────
+async function loadPatients() {
+  show('pLoading'); hide('pEmpty'); document.getElementById('pGrid').innerHTML = '';
+  try {
+    const d = await fetch('/api/hospital/patients').then(r=>r.json());
+    renderGrid(d.patients||[], 'pGrid', 'pEmpty');
+  } catch(e) { show('pEmpty'); }
+  hide('pLoading');
+}
+
+async function doSearch() {
+  const cond = document.getElementById('searchInput').value.trim();
+  const ic   = document.getElementById('searchIncludeConnected').checked ? '1' : '0';
+  hideAuto();
+  show('sLoading'); hide('sEmpty'); document.getElementById('sGrid').innerHTML = '';
+  try {
+    const url = '/api/hospital/patients?condition='+encodeURIComponent(cond)+'&include_connected='+ic;
+    const d   = await fetch(url).then(r=>r.json());
+    renderGrid(d.patients||[], 'sGrid', 'sEmpty');
+  } catch(e) { show('sEmpty'); }
+  hide('sLoading');
+}
+
+function renderGrid(patients, gridId, emptyId) {
+  const grid = document.getElementById(gridId);
+  if (!patients.length) { show(emptyId); return; }
+  hide(emptyId);
+  patients.forEach(p => {
+    const age  = p.dob ? Math.floor((Date.now()-new Date(p.dob))/31557600000) : '?';
+    const conds= (p.conditions||[]).slice(0,5);
+    const col  = document.createElement('div');
+    col.className = 'col-xl-3 col-lg-4 col-md-6';
+
+    const card = document.createElement('div');
+    card.className = 'patient-card d-flex flex-column';
+
+    const name = document.createElement('h6');
+    name.className = 'mb-1';
+    name.textContent = (p.first_name||'') + ' ' + (p.last_name||'');
+
+    const meta = document.createElement('div');
+    meta.className = 'text-muted small mb-2';
+    meta.textContent = `Age ${age} • ${p.gender||'?'}`;
+
+    const condWrap = document.createElement('div');
+    condWrap.className = 'flex-grow-1 mb-3';
+    conds.forEach(c => {
+      const badge = document.createElement('span');
+      badge.className = 'badge bg-secondary me-1 mb-1';
+      badge.textContent = c;
+      condWrap.appendChild(badge);
+    });
+    if ((p.conditions||[]).length > 5) {
+      const more = document.createElement('span');
+      more.className = 'text-muted small';
+      more.textContent = `+${p.conditions.length-5} more`;
+      condWrap.appendChild(more);
+    }
+
+    const btn = document.createElement('button');
+    btn.className = 'btn btn-sm btn-outline-success w-100';
+    btn.textContent = 'Connect with Patient';
+    btn.addEventListener('click', () => openConnectModal(p.id, name.textContent));
+
+    card.appendChild(name); card.appendChild(meta);
+    card.appendChild(condWrap); card.appendChild(btn);
+    col.appendChild(card);
+    grid.appendChild(col);
+  });
+}
+
+// ── CONNECT WITH PATIENT ──────────────────────────────────────────────────────
+function openConnectModal(patientId, name) {
+  targetPatientId = patientId;
+  document.getElementById('modalPatientInfo').textContent = 'Patient: ' + name;
+  document.getElementById('modalMsg').value = '';
+  connectModal.show();
+}
+
+async function submitConnect() {
+  if (!targetPatientId) return;
+  try {
+    const r = await fetch('/api/hospital/connect', {
+      method:'POST', headers:{'Content-Type':'application/json'},
+      body: JSON.stringify({patient_id:targetPatientId, message:document.getElementById('modalMsg').value.trim()})
+    });
+    const d = await r.json();
+    if (r.ok) { connectModal.hide(); switchTab('connections'); }
+    else alert(d.error||'Failed');
+  } catch(e) { alert('Network error'); }
+}
+
+// ── CONNECTIONS ───────────────────────────────────────────────────────────────
+async function loadConnections() {
+  try {
+    const d = await fetch('/api/hospital/connections').then(r=>r.json());
+    const conns = d.connections||[];
+    const empty = document.getElementById('cEmpty');
+    const wrap  = document.getElementById('cWrap');
+    const tbody = document.getElementById('cTbody');
+    if (!conns.length) { show('cEmpty'); hide('cWrap'); return; }
+    hide('cEmpty'); show('cWrap');
+    tbody.innerHTML = '';
+    conns.forEach(c => {
+      const age  = c.dob ? Math.floor((Date.now()-new Date(c.dob))/31557600000) : '?';
+      const cstr = (Array.isArray(c.conditions)?c.conditions:[]).slice(0,3).join(', ');
+      const sc   = {pending:'st-pending',accepted:'st-accepted',rejected:'st-rejected',completed:'st-completed'}[c.status]||'';
+      const pname = (c.first_name||'')+' '+(c.last_name||'');
+      const tr   = document.createElement('tr');
+      tr.innerHTML = [
+        escH(pname),
+        escH(age+' / '+(c.gender||'?')),
+        escH(cstr||'—'),
+        escH((c.trial_title||'—').substring(0,40)),
+        escH(c.initiated_by||''),
+        `<span class="${sc} fw-bold">${escH(c.status)}</span>`,
+        escH((c.message||'').substring(0,50)),
+        statusBtns(c.id, c.status),
+        `<button class="btn btn-sm btn-outline-info py-0 px-2" style="font-size:.72rem"
+           data-cid="${escA(c.id)}" data-chat="${escA(pname)}">💬 Chat</button>`,
+      ].map(v=>`<td>${v}</td>`).join('');
+      tbody.appendChild(tr);
+    });
+  } catch(e) {}
+}
+
+function statusBtns(cid, status) {
+  const sa = escA(cid);
+  if (status === 'pending') return `
+    <div class="d-flex gap-1">
+      <button class="btn btn-sm btn-success py-0 px-2" style="font-size:.72rem" data-cid="${sa}" data-st="accepted">Accept</button>
+      <button class="btn btn-sm btn-danger py-0 px-2" style="font-size:.72rem" data-cid="${sa}" data-st="rejected">Reject</button>
+    </div>`;
+  if (status === 'accepted') return `
+    <button class="btn btn-sm btn-secondary py-0 px-2" style="font-size:.72rem" data-cid="${sa}" data-st="completed">Complete</button>`;
+  return '';
+}
+
+async function updateStatus(cid, status) {
+  await fetch(`/api/hospital/connections/${cid}/status`, {
+    method:'PUT', headers:{'Content-Type':'application/json'},
+    body: JSON.stringify({status})
+  });
+  loadConnections();
+}
+
+// ── MY TRIALS ────────────────────────────────────────────────────────────────
+async function loadTrials() {
+  show('tLoading'); hide('tEmpty'); document.getElementById('tGrid').innerHTML = '';
+  try {
+    const d = await fetch('/api/hospital/trials').then(r => r.json());
+    const trials = d.trials || [];
+    if (!trials.length) { show('tEmpty'); hide('tLoading'); return; }
+    trials.forEach(t => {
+      const stc     = t.status === 'RECRUITING' ? 'success' : t.status.includes('ACTIVE') ? 'info' : 'secondary';
+      const tierCls = t.match_tier === 1 ? 'bg-success' : 'bg-secondary';
+      const div = document.createElement('div');
+      div.className = 'col-12';
+      div.innerHTML = `<div class="patient-card mb-2">
+        <div class="d-flex flex-wrap gap-1 align-items-center mb-2">
+          <span class="badge bg-${stc}">${escH(t.status)}</span>
+          ${t.phase&&t.phase!=='N/A'?`<span class="badge bg-info text-dark">${escH(t.phase)}</span>`:''}
+          <span class="badge ${tierCls} ms-auto">${escH(t.match_reason)}</span>
+        </div>
+        <h6 class="mb-1">${escH(t.title)}</h6>
+        <div class="text-muted small mb-2">${escH(t.trial_id)}</div>
+        ${t.summary?`<p class="text-muted mb-2" style="font-size:.8rem">${escH(t.summary.substring(0,260)+(t.summary.length>260?'…':''))}</p>`:''}
+        <div class="d-flex flex-wrap gap-3 small text-muted mb-2">
+          ${t.facility_name?`<span>📍 ${escH(t.facility_name)}</span>`:''}
+          ${t.location?`<span>🗺 ${escH(t.location)}</span>`:''}
+          ${t.n_sites?`<span>🏥 ${t.n_sites} site${t.n_sites!==1?'s':''}</span>`:''}
+          ${t.enrollment?`<span>👥 ${t.enrollment}</span>`:''}
+          <span>Ages ${t.min_age}–${t.max_age} · ${escH(t.sex||'ALL')}</span>
+        </div>
+        <a href="https://clinicaltrials.gov/study/${escA(t.trial_id)}" target="_blank" rel="noopener"
+           class="btn btn-sm btn-outline-secondary">View on ClinicalTrials.gov ↗</a>
+      </div>`;
+      document.getElementById('tGrid').appendChild(div);
+    });
+  } catch(e) { show('tEmpty'); }
+  hide('tLoading');
+}
+
+// ── MESSAGING ─────────────────────────────────────────────────────────────────
+async function openMsgModal(connId, label) {
+  currentConnId = connId;
+  document.getElementById('msgModalTitle').textContent = 'Chat — ' + label;
+  document.getElementById('msgThread').innerHTML =
+    '<div class="text-center text-muted small py-4">Loading…</div>';
+  document.getElementById('msgInput').value = '';
+  msgModal.show();
+  await loadMsgs();
+}
+
+async function loadMsgs() {
+  try {
+    const d = await fetch('/api/hospital/connections/'+encodeURIComponent(currentConnId)+'/messages').then(r=>r.json());
+    renderMsgs(d.messages || []);
+  } catch(e) {
+    document.getElementById('msgThread').innerHTML =
+      '<div class="text-danger small text-center py-4">Error loading messages.</div>';
+  }
+}
+
+function renderMsgs(msgs) {
+  const thread = document.getElementById('msgThread');
+  if (!msgs.length) {
+    thread.innerHTML = '<div class="text-muted small text-center py-4">No messages yet. Start the conversation.</div>';
+    return;
+  }
+  thread.innerHTML = '';
+  msgs.forEach(m => {
+    const mine = m.sender_role === 'hospital';
+    const wrap = document.createElement('div');
+    wrap.className = 'd-flex mb-2 ' + (mine ? 'justify-content-end' : 'justify-content-start');
+    const bubble = document.createElement('div');
+    bubble.style.cssText = 'max-width:72%;padding:.5rem .75rem;border-radius:12px;font-size:.85rem;'
+      + (mine ? 'background:#1f6feb;color:#fff;' : 'background:#21262d;color:#c9d1d9;');
+    const bodyEl = document.createElement('div'); bodyEl.textContent = m.body;
+    const ts = document.createElement('div');
+    ts.style.cssText = 'font-size:.65rem;opacity:.6;margin-top:.2rem;text-align:right';
+    ts.textContent = (m.created_at||'').replace('T',' ').substring(0,16);
+    bubble.appendChild(bodyEl); bubble.appendChild(ts);
+    wrap.appendChild(bubble); thread.appendChild(wrap);
+  });
+  thread.scrollTop = thread.scrollHeight;
+}
+
+async function sendMsg() {
+  const input = document.getElementById('msgInput');
+  const body  = input.value.trim();
+  if (!body || !currentConnId) return;
+  input.value = '';
+  try {
+    await fetch('/api/hospital/connections/'+encodeURIComponent(currentConnId)+'/messages', {
+      method:'POST', headers:{'Content-Type':'application/json'},
+      body: JSON.stringify({body})
+    });
+    await loadMsgs();
+  } catch(e) {}
+}
+
+// ── INBOX ─────────────────────────────────────────────────────────────────────
+async function loadInbox() {
+  document.getElementById('inboxLoading').style.display = '';
+  document.getElementById('inboxEmpty').style.display   = 'none';
+  document.getElementById('inboxList').innerHTML = '';
+  try {
+    const d = await fetch('/api/hospital/inbox').then(r => r.json());
+    document.getElementById('inboxLoading').style.display = 'none';
+    const threads = d.threads || [];
+    if (!threads.length) { document.getElementById('inboxEmpty').style.display = ''; return; }
+    let totalUnread = 0;
+    threads.forEach(t => {
+      totalUnread += t.unread_count || 0;
+      const pname   = (t.first_name || '') + ' ' + (t.last_name || '');
+      const trial   = (t.trial_title || 'General Inquiry').substring(0, 60);
+      const preview = (t.last_message || 'No messages yet').substring(0, 100);
+      const card = document.createElement('div');
+      card.className = 'card bg-secondary mb-2';
+      const body = document.createElement('div');
+      body.className = 'card-body d-flex justify-content-between align-items-center gap-3';
+      const info = document.createElement('div');
+      info.style.minWidth = '0';
+      const nameEl = document.createElement('strong'); nameEl.textContent = pname;
+      info.appendChild(nameEl);
+      if (t.unread_count) {
+        const bEl = document.createElement('span');
+        bEl.className = 'badge bg-danger ms-2'; bEl.textContent = t.unread_count;
+        info.appendChild(bEl);
+      }
+      const trialEl = document.createElement('div');
+      trialEl.className = 'text-muted small text-truncate'; trialEl.textContent = trial;
+      const previewEl = document.createElement('div');
+      previewEl.className = 'text-secondary small fst-italic text-truncate'; previewEl.textContent = preview;
+      info.appendChild(trialEl); info.appendChild(previewEl);
+      const btn = document.createElement('button');
+      btn.className = 'btn btn-sm btn-outline-light flex-shrink-0';
+      btn.textContent = 'Open';
+      btn.dataset.cid = t.id; btn.dataset.chat = pname;
+      body.appendChild(info); body.appendChild(btn);
+      card.appendChild(body);
+      document.getElementById('inboxList').appendChild(card);
+    });
+    const badge = document.getElementById('inboxBadge');
+    if (totalUnread > 0) { badge.textContent = totalUnread; badge.style.display = ''; }
+    else badge.style.display = 'none';
+  } catch(e) {
+    document.getElementById('inboxLoading').style.display = 'none';
+    document.getElementById('inboxEmpty').style.display = '';
+  }
+}
+
+// ── CONDITION AUTOCOMPLETE ────────────────────────────────────────────────────
+async function condAuto(q) {
+  if (q.length < 2) { hideAuto(); return; }
+  try {
+    const d = await fetch('/api/conditions/autocomplete?q='+encodeURIComponent(q)).then(r=>r.json());
+    const list = document.getElementById('searchAutoList');
+    list.innerHTML = '';
+    if (!(d.results||[]).length) { list.style.display='none'; return; }
+    d.results.forEach(item => {
+      const li = document.createElement('li');
+      li.textContent = item;
+      li.addEventListener('mousedown', e => {
+        e.preventDefault();
+        document.getElementById('searchInput').value = item;
+        hideAuto(); doSearch();
+      });
+      list.appendChild(li);
+    });
+    list.style.display = 'block';
+  } catch(e) { hideAuto(); }
+}
+function hideAuto() { document.getElementById('searchAutoList').style.display = 'none'; }
+
+// ── UTILS ─────────────────────────────────────────────────────────────────────
+function show(id) { document.getElementById(id).style.display = ''; }
+function hide(id) { document.getElementById(id).style.display = 'none'; }
+function escH(s) {
+  const d = document.createElement('div');
+  d.appendChild(document.createTextNode(String(s||'')));
+  return d.innerHTML;
+}
+function escA(s) {
+  return String(s||'').replace(/&/g,'&amp;').replace(/"/g,'&quot;').replace(/</g,'&lt;').replace(/>/g,'&gt;');
+}
+async function logout() {
+  await fetch('/auth/logout', {method:'POST'}); location.href = '/';
+}
+</script>
+</body>
+</html>

templates/landing.html

@@ -0,0 +1,202 @@
+<!DOCTYPE html>
+<html lang="en" data-bs-theme="dark">
+<head>
+<meta charset="UTF-8">
+<meta name="viewport" content="width=device-width, initial-scale=1.0">
+<title>Second Life — Clinical Trial Matching</title>
+<link href="https://cdn.jsdelivr.net/npm/bootstrap@5.3.2/dist/css/bootstrap.min.css" rel="stylesheet">
+<style>
+body{background:linear-gradient(135deg,#0d1117 0%,#161b22 100%);min-height:100vh;}
+.hero-title{font-size:3rem;font-weight:800;background:linear-gradient(90deg,#58a6ff,#79c0ff);-webkit-background-clip:text;-webkit-text-fill-color:transparent;background-clip:text;}
+.portal-card{background:#161b22;border:1px solid #30363d;border-radius:12px;padding:2rem;}
+.form-control,.form-select{background:#0d1117;border-color:#30363d;color:#e6edf3;}
+.form-control:focus,.form-select:focus{background:#0d1117;border-color:#58a6ff;color:#e6edf3;box-shadow:0 0 0 .2rem rgba(88,166,255,.25);}
+.nav-tabs .nav-link{color:#8b949e;border-color:transparent;}
+.nav-tabs .nav-link.active{color:#e6edf3;background:#161b22;border-color:#30363d #30363d #161b22;}
+.err{color:#f85149;font-size:.85rem;min-height:1.2rem;}
+</style>
+</head>
+<body>
+<div class="container py-5">
+  <div class="text-center mb-5">
+    <div class="text-primary mb-2" style="font-size:2rem">⬡</div>
+    <h1 class="hero-title">Second Life</h1>
+    <p class="text-secondary fs-5 mb-1">AI-Powered Clinical Trial Matching</p>
+    <p class="text-muted small">Connecting patients with life-changing research opportunities</p>
+  </div>
+
+  <div class="row g-4 justify-content-center">
+    <!-- Patient Portal -->
+    <div class="col-lg-5">
+      <div class="portal-card">
+        <div class="text-center mb-3" style="font-size:2.5rem">🧑‍⚕️</div>
+        <h4 class="text-center mb-1">Patient Portal</h4>
+        <p class="text-muted text-center small mb-3">Find clinical trials matched to your conditions</p>
+        <ul class="nav nav-tabs mb-3" id="patientTabs">
+          <li class="nav-item"><a class="nav-link active" href="#" data-pt="p-login">Login</a></li>
+          <li class="nav-item"><a class="nav-link" href="#" data-pt="p-register">Register</a></li>
+        </ul>
+        <!-- Login -->
+        <div id="p-login">
+          <input type="text" class="form-control mb-2" id="pl-user" placeholder="Username">
+          <input type="password" class="form-control mb-3" id="pl-pass" placeholder="Password">
+          <div class="err mb-2" id="pl-err"></div>
+          <button class="btn btn-primary w-100" onclick="patientLogin()">Login as Patient</button>
+          <p class="text-muted text-center mt-2 small">Demo: john_doe / pass123</p>
+        </div>
+        <!-- Register -->
+        <div id="p-register" style="display:none">
+          <div class="row g-2 mb-2">
+            <div class="col"><input type="text" class="form-control form-control-sm" id="pr-fn" placeholder="First Name"></div>
+            <div class="col"><input type="text" class="form-control form-control-sm" id="pr-ln" placeholder="Last Name"></div>
+          </div>
+          <div class="row g-2 mb-2">
+            <div class="col"><input type="date" class="form-control form-control-sm" id="pr-dob"></div>
+            <div class="col">
+              <select class="form-select form-select-sm" id="pr-gender">
+                <option value="">Gender</option>
+                <option value="M">Male</option>
+                <option value="F">Female</option>
+                <option value="O">Other</option>
+              </select>
+            </div>
+          </div>
+          <input type="text" class="form-control form-control-sm mb-2" id="pr-addr" placeholder="Address (123 Main St Boston MA 02101)">
+          <input type="text" class="form-control form-control-sm mb-2" id="pr-user" placeholder="Username">
+          <input type="password" class="form-control form-control-sm mb-3" id="pr-pass" placeholder="Password">
+          <div class="err mb-2" id="pr-err"></div>
+          <button class="btn btn-primary w-100" onclick="patientRegister()">Create Patient Account</button>
+        </div>
+      </div>
+    </div>
+
+    <!-- Hospital Portal -->
+    <div class="col-lg-5">
+      <div class="portal-card">
+        <div class="text-center mb-3" style="font-size:2.5rem">🏥</div>
+        <h4 class="text-center mb-1">Hospital / Research Portal</h4>
+        <p class="text-muted text-center small mb-3">Find eligible patients for your clinical trials</p>
+        <ul class="nav nav-tabs mb-3" id="hospitalTabs">
+          <li class="nav-item"><a class="nav-link active" href="#" data-ht="h-login">Login</a></li>
+          <li class="nav-item"><a class="nav-link" href="#" data-ht="h-register">Register</a></li>
+        </ul>
+        <!-- Login -->
+        <div id="h-login">
+          <input type="text" class="form-control mb-2" id="hl-user" placeholder="Username">
+          <input type="password" class="form-control mb-3" id="hl-pass" placeholder="Password">
+          <div class="err mb-2" id="hl-err"></div>
+          <button class="btn btn-success w-100" onclick="hospitalLogin()">Login as Hospital</button>
+          <p class="text-muted text-center mt-2 small">Demo: mgh / mgh123</p>
+        </div>
+        <!-- Register -->
+        <div id="h-register" style="display:none">
+          <input type="text" class="form-control form-control-sm mb-2" id="hr-name" placeholder="Hospital / Institution Name">
+          <input type="text" class="form-control form-control-sm mb-2" id="hr-loc" placeholder="Location (City, ST) e.g. Boston, MA">
+          <input type="text" class="form-control form-control-sm mb-2" id="hr-conds"
+                 placeholder="Research conditions (comma-separated, e.g. diabetes, hypertension)">
+          <input type="text" class="form-control form-control-sm mb-2" id="hr-user" placeholder="Username">
+          <input type="password" class="form-control form-control-sm mb-3" id="hr-pass" placeholder="Password">
+          <div class="err mb-2" id="hr-err"></div>
+          <button class="btn btn-success w-100" onclick="hospitalRegister()">Create Hospital Account</button>
+        </div>
+      </div>
+    </div>
+  </div>
+
+  <p class="text-center text-muted small mt-4">DSCI 5260 | Group 7 — Second Life Clinical Trial Matching</p>
+</div>
+
+<script src="https://cdn.jsdelivr.net/npm/bootstrap@5.3.2/dist/js/bootstrap.bundle.min.js"></script>
+<script>
+// Enter key submit helpers
+function _onEnter(inputId, fn) {
+  document.getElementById(inputId).addEventListener('keydown', e => { if(e.key==='Enter') fn(); });
+}
+document.addEventListener('DOMContentLoaded', () => {
+  _onEnter('pl-pass',   patientLogin);
+  _onEnter('pl-user',   patientLogin);
+  _onEnter('pr-pass',   patientRegister);
+  _onEnter('hl-pass',   hospitalLogin);
+  _onEnter('hl-user',   hospitalLogin);
+  _onEnter('hr-pass',   hospitalRegister);
+});
+
+// Tab switching
+document.querySelectorAll('#patientTabs .nav-link').forEach(a => {
+  a.addEventListener('click', e => {
+    e.preventDefault();
+    document.querySelectorAll('#patientTabs .nav-link').forEach(x => x.classList.remove('active'));
+    a.classList.add('active');
+    document.getElementById('p-login').style.display    = a.dataset.pt === 'p-login'    ? '' : 'none';
+    document.getElementById('p-register').style.display = a.dataset.pt === 'p-register' ? '' : 'none';
+  });
+});
+document.querySelectorAll('#hospitalTabs .nav-link').forEach(a => {
+  a.addEventListener('click', e => {
+    e.preventDefault();
+    document.querySelectorAll('#hospitalTabs .nav-link').forEach(x => x.classList.remove('active'));
+    a.classList.add('active');
+    document.getElementById('h-login').style.display    = a.dataset.ht === 'h-login'    ? '' : 'none';
+    document.getElementById('h-register').style.display = a.dataset.ht === 'h-register' ? '' : 'none';
+  });
+});
+
+async function patientLogin() {
+  const u = document.getElementById('pl-user').value.trim();
+  const p = document.getElementById('pl-pass').value;
+  const e = document.getElementById('pl-err');
+  e.textContent = '';
+  if (!u || !p) { e.textContent = 'Enter username and password'; return; }
+  const r = await post('/auth/patient/login', {username:u, password:p});
+  if (r.ok) location.href = '/patient'; else e.textContent = (await r.json()).error || 'Login failed';
+}
+
+async function patientRegister() {
+  const body = {
+    username:   document.getElementById('pr-user').value.trim(),
+    password:   document.getElementById('pr-pass').value,
+    first_name: document.getElementById('pr-fn').value.trim(),
+    last_name:  document.getElementById('pr-ln').value.trim(),
+    dob:        document.getElementById('pr-dob').value,
+    gender:     document.getElementById('pr-gender').value,
+    address:    document.getElementById('pr-addr').value.trim(),
+  };
+  const e = document.getElementById('pr-err');
+  e.textContent = '';
+  if (!body.username || !body.password) { e.textContent = 'Username and password required'; return; }
+  const r = await post('/auth/patient/register', body);
+  if (r.ok) location.href = '/patient'; else e.textContent = (await r.json()).error || 'Registration failed';
+}
+
+async function hospitalLogin() {
+  const u = document.getElementById('hl-user').value.trim();
+  const p = document.getElementById('hl-pass').value;
+  const e = document.getElementById('hl-err');
+  e.textContent = '';
+  if (!u || !p) { e.textContent = 'Enter username and password'; return; }
+  const r = await post('/auth/hospital/login', {username:u, password:p});
+  if (r.ok) location.href = '/hospital'; else e.textContent = (await r.json()).error || 'Login failed';
+}
+
+async function hospitalRegister() {
+  const condsRaw = document.getElementById('hr-conds').value.trim();
+  const body = {
+    username:             document.getElementById('hr-user').value.trim(),
+    password:             document.getElementById('hr-pass').value,
+    hospital_name:        document.getElementById('hr-name').value.trim(),
+    location:             document.getElementById('hr-loc').value.trim(),
+    research_conditions:  condsRaw ? condsRaw.split(',').map(s => s.trim().toLowerCase()).filter(Boolean) : [],
+  };
+  const e = document.getElementById('hr-err');
+  e.textContent = '';
+  if (!body.username || !body.password || !body.hospital_name) { e.textContent = 'All fields required'; return; }
+  const r = await post('/auth/hospital/register', body);
+  if (r.ok) location.href = '/hospital'; else e.textContent = (await r.json()).error || 'Registration failed';
+}
+
+function post(url, body) {
+  return fetch(url, {method:'POST', headers:{'Content-Type':'application/json'}, body:JSON.stringify(body)});
+}
+</script>
+</body>
+</html>

templates/patient.html

@@ -0,0 +1,866 @@
+<!DOCTYPE html>
+<html lang="en" data-bs-theme="dark">
+<head>
+<meta charset="UTF-8">
+<meta name="viewport" content="width=device-width, initial-scale=1.0">
+<title>Patient Portal — Second Life</title>
+<link href="https://cdn.jsdelivr.net/npm/bootstrap@5.3.2/dist/css/bootstrap.min.css" rel="stylesheet">
+<style>
+body{background:#0d1117;}
+.trial-card{background:#161b22;border:1px solid #30363d;border-radius:10px;padding:1.5rem;}
+.score-circle{width:72px;height:72px;border-radius:50%;border:3px solid currentColor;display:flex;flex-direction:column;align-items:center;justify-content:center;font-weight:700;font-size:1.1rem;line-height:1.1;flex-shrink:0;}
+.score-circle small{font-size:.6rem;font-weight:500;}
+.score-high{color:#2ea043;}.score-mid{color:#d29922;}.score-low{color:#8b949e;}
+.detail-grid{display:grid;grid-template-columns:repeat(4,1fr);gap:.6rem;background:#0d1117;border-radius:8px;padding:.75rem;}
+@media(max-width:576px){.detail-grid{grid-template-columns:repeat(2,1fr);}}
+.detail-grid .lbl{font-size:.7rem;color:#8b949e;text-transform:uppercase;letter-spacing:.05em;}
+.detail-grid .val{font-size:.9rem;font-weight:600;}
+.detail-grid .col-span-2{grid-column:span 2;}
+.bar-section{background:#0d1117;border-radius:8px;padding:.75rem;}
+.bar-lbl{font-size:.75rem;color:#8b949e;}.bar-val{font-size:.75rem;font-weight:600;}
+.progress{height:5px;background:#21262d;}
+.sec-head{font-size:.7rem;text-transform:uppercase;letter-spacing:.08em;color:#8b949e;margin-bottom:.5rem;font-weight:600;}
+.autocomplete-list{position:absolute;z-index:1000;width:100%;background:#161b22;border:1px solid #30363d;border-radius:0 0 6px 6px;max-height:190px;overflow-y:auto;display:none;list-style:none;padding:0;margin:0;}
+.autocomplete-list li{padding:.3rem .6rem;cursor:pointer;font-size:.85rem;}
+.autocomplete-list li:hover{background:#21262d;}
+.criteria-pre{white-space:pre-wrap;font-family:inherit;font-size:.78rem;background:#0d1117;padding:.6rem;border-radius:4px;max-height:200px;overflow-y:auto;}
+.st-pending{color:#d29922;}.st-accepted{color:#2ea043;}.st-rejected{color:#f85149;}.st-completed{color:#58a6ff;}
+.form-control,.form-select{background:#0d1117 !important;border-color:#30363d;color:#e6edf3 !important;}
+.form-control:focus,.form-select:focus{border-color:#58a6ff !important;box-shadow:0 0 0 .2rem rgba(88,166,255,.2) !important;}
+</style>
+</head>
+<body>
+<!-- Navbar -->
+<nav class="navbar border-bottom border-secondary px-3 py-2" style="background:#161b22">
+  <span class="navbar-brand fw-bold" style="color:#58a6ff">⬡ Second Life</span>
+  <div class="d-flex align-items-center gap-3">
+    <span class="text-light small">{{ patient.first_name }} {{ patient.last_name }}</span>
+    <span class="badge bg-primary">Patient</span>
+    <button class="btn btn-sm btn-outline-secondary" id="logoutBtn">Logout</button>
+  </div>
+</nav>
+<!-- System banner -->
+<div id="sysBanner" class="alert alert-warning rounded-0 mb-0 py-2 text-center small" style="display:none">
+  ⏳ Pipeline still loading — trial matches will be available once model training completes.
+</div>
+
+<div class="container-xl py-3">
+  <ul class="nav nav-tabs mb-4" id="mainTabs">
+    <li class="nav-item"><button class="nav-link active" data-tab="profile">My Profile</button></li>
+    <li class="nav-item"><button class="nav-link" data-tab="trials">Suggested Trials</button></li>
+    <li class="nav-item"><button class="nav-link" data-tab="connections">My Connections</button></li>
+    <li class="nav-item"><button class="nav-link" data-tab="inbox">Inbox <span id="inboxBadge" class="badge bg-danger ms-1" style="display:none">0</span></button></li>
+  </ul>
+
+  <!-- ===== PROFILE TAB ===== -->
+  <div id="tab-profile">
+    <div class="row g-4">
+      <div class="col-lg-7">
+        <div class="card border-secondary" style="background:#161b22">
+          <div class="card-header border-secondary">Personal Information</div>
+          <div class="card-body">
+            <div class="row g-3">
+              <div class="col-sm-6">
+                <label class="form-label small">First Name</label>
+                <input type="text" class="form-control form-control-sm" id="pf-fn">
+              </div>
+              <div class="col-sm-6">
+                <label class="form-label small">Last Name</label>
+                <input type="text" class="form-control form-control-sm" id="pf-ln">
+              </div>
+              <div class="col-sm-6">
+                <label class="form-label small">Date of Birth</label>
+                <input type="date" class="form-control form-control-sm" id="pf-dob">
+              </div>
+              <div class="col-sm-6">
+                <label class="form-label small">Gender</label>
+                <select class="form-select form-select-sm" id="pf-gender">
+                  <option value="">Select</option>
+                  <option value="M">Male</option>
+                  <option value="F">Female</option>
+                  <option value="O">Other</option>
+                </select>
+              </div>
+              <div class="col-12">
+                <label class="form-label small">Address</label>
+                <input type="text" class="form-control form-control-sm" id="pf-addr" placeholder="123 Main St Boston MA 02101">
+              </div>
+            </div>
+          </div>
+        </div>
+
+        <div class="card border-secondary mt-3" style="background:#161b22">
+          <div class="card-header border-secondary d-flex justify-content-between">
+            <span>Medical Conditions</span>
+            <small class="text-muted">Used for trial matching</small>
+          </div>
+          <div class="card-body">
+            <div id="condTags" class="d-flex flex-wrap gap-1 mb-2" style="min-height:1.8rem"></div>
+            <div class="position-relative">
+              <input type="text" class="form-control form-control-sm" id="condInput"
+                     placeholder="Type to search and add conditions…" autocomplete="off">
+              <ul class="autocomplete-list" id="condAutoList"></ul>
+            </div>
+          </div>
+        </div>
+
+        <div class="card border-secondary mt-3" style="background:#161b22">
+          <div class="card-header border-secondary">Medications</div>
+          <div class="card-body">
+            <div id="medTags" class="d-flex flex-wrap gap-1 mb-2" style="min-height:1.8rem"></div>
+            <input type="text" class="form-control form-control-sm" id="medInput"
+                   placeholder="Type medication name and press Enter…">
+          </div>
+        </div>
+
+        <div class="card border-secondary mt-3" style="background:#161b22">
+          <div class="card-header border-secondary d-flex justify-content-between">
+            <span>Documents</span>
+            <small class="text-muted">PDF, DOCX, TXT, PNG, JPG — max 10 MB</small>
+          </div>
+          <div class="card-body">
+            <div id="docList" class="mb-2"></div>
+            <div class="d-flex gap-2">
+              <input type="file" class="form-control form-control-sm" id="docFileInput"
+                     accept=".pdf,.docx,.doc,.txt,.png,.jpg,.jpeg" style="flex:1">
+              <button class="btn btn-sm btn-outline-primary" id="docUploadBtn">Upload</button>
+            </div>
+            <div id="docMsg" class="small mt-1" style="min-height:1rem"></div>
+          </div>
+        </div>
+      </div>
+
+      <div class="col-lg-5">
+        <div class="card border-secondary" style="background:#161b22">
+          <div class="card-header border-secondary">Trial Participation</div>
+          <div class="card-body">
+            <div class="form-check form-switch mb-3">
+              <input class="form-check-input" type="checkbox" id="openToTrials">
+              <label class="form-check-label" for="openToTrials">
+                <strong>Open to Clinical Trials</strong>
+                <div class="text-muted small">Allow hospitals to see your basic profile when searching for eligible patients.</div>
+              </label>
+            </div>
+            <div class="alert alert-info py-2 small">
+              🔒 Your detailed information is only shared when <strong>you</strong> click "Connect with Hospital."
+            </div>
+          </div>
+        </div>
+        <div class="card border-secondary mt-3" style="background:#161b22">
+          <div class="card-header border-secondary">Account</div>
+          <div class="card-body small text-muted" id="acctInfo"></div>
+        </div>
+        <button class="btn btn-primary w-100 mt-3" id="saveBtn">Save Profile</button>
+        <div id="saveMsg" class="text-center small mt-2" style="min-height:1.2rem"></div>
+      </div>
+    </div>
+  </div>
+
+  <!-- ===== TRIALS TAB ===== -->
+  <div id="tab-trials" style="display:none">
+    <div class="d-flex justify-content-between align-items-center mb-3">
+      <div>
+        <h6 class="mb-0">Trial Recommendations</h6>
+        <small class="text-muted">AI-ranked by eligibility probability using your profile data</small>
+      </div>
+      <button class="btn btn-primary btn-sm" id="getMatchesBtn">Get My Matches</button>
+    </div>
+    <div id="trialsLoading" class="text-center py-5" style="display:none">
+      <div class="spinner-border text-primary"></div>
+      <p class="text-muted mt-2 small">Analyzing your profile against clinical trials…</p>
+    </div>
+    <div id="trialsMsg" class="text-center py-5 text-muted" style="display:none"></div>
+    <div class="alert alert-secondary py-2 small mb-3" id="trialsDisclaimer" style="display:none">
+      <strong>About these scores:</strong> Match percentages are predictions from a model trained on
+      synthetic patient data and rule-based labels — not validated clinical eligibility determinations.
+      Always consult a healthcare provider before enrolling in any trial.
+    </div>
+    <div id="trialsResults"></div>
+  </div>
+
+  <!-- ===== CONNECTIONS TAB ===== -->
+  <div id="tab-connections" style="display:none">
+    <div class="d-flex justify-content-between align-items-center mb-3">
+      <h6 class="mb-0">My Connections</h6>
+      <button class="btn btn-sm btn-outline-secondary" id="refreshConnsBtn">Refresh</button>
+    </div>
+    <div id="connsMsg" class="text-center py-5 text-muted" style="display:none">
+      No connections yet. Find a trial and click "Connect with Hospital."
+    </div>
+    <div id="connsWrap" style="display:none">
+      <div class="table-responsive">
+        <table class="table table-dark table-hover table-sm align-middle">
+          <thead><tr>
+            <th>Hospital</th><th>Trial</th><th>Initiated By</th>
+            <th>Status</th><th>Message</th><th>Date</th><th>Chat</th>
+          </tr></thead>
+          <tbody id="connsTbody"></tbody>
+        </table>
+      </div>
+    </div>
+  </div>
+
+  <!-- ===== INBOX TAB ===== -->
+  <div id="tab-inbox" style="display:none">
+    <div class="d-flex justify-content-between align-items-center mb-3">
+      <h6 class="mb-0">Inbox</h6>
+      <button class="btn btn-sm btn-outline-secondary" onclick="loadInbox()">Refresh</button>
+    </div>
+    <div id="inboxLoading" class="text-center text-muted py-4">Loading…</div>
+    <div id="inboxEmpty" class="text-center text-muted py-4" style="display:none">No message threads yet. Connect with a hospital to start messaging.</div>
+    <div id="inboxList"></div>
+  </div>
+</div>
+
+<!-- Messages Modal -->
+<div class="modal fade" id="msgModal" tabindex="-1">
+  <div class="modal-dialog modal-lg">
+    <div class="modal-content border-secondary" style="background:#161b22">
+      <div class="modal-header border-secondary">
+        <h5 class="modal-title" id="msgModalTitle">Messages</h5>
+        <button type="button" class="btn-close btn-close-white" data-bs-dismiss="modal"></button>
+      </div>
+      <div class="modal-body p-0">
+        <div id="msgThread" style="height:320px;overflow-y:auto;padding:1rem;background:#0d1117"></div>
+        <div class="p-3 border-top border-secondary d-flex gap-2">
+          <input type="text" class="form-control form-control-sm" id="msgInput"
+                 placeholder="Type a message…" maxlength="2000">
+          <button class="btn btn-primary btn-sm px-3" id="msgSendBtn">Send</button>
+        </div>
+      </div>
+    </div>
+  </div>
+</div>
+
+<!-- Connect with Hospital Modal -->
+<div class="modal fade" id="connectModal" tabindex="-1">
+  <div class="modal-dialog">
+    <div class="modal-content border-secondary" style="background:#161b22">
+      <div class="modal-header border-secondary">
+        <h5 class="modal-title">Connect with Hospital</h5>
+        <button type="button" class="btn-close btn-close-white" data-bs-dismiss="modal"></button>
+      </div>
+      <div class="modal-body">
+        <p class="text-muted small" id="modalTrialInfo"></p>
+        <div id="hospitalList" class="mb-3" style="max-height:420px;overflow-y:auto"></div>
+        <label class="form-label small">Message <span class="text-muted">(optional)</span></label>
+        <textarea class="form-control form-control-sm" id="modalMsg" rows="3"
+                  placeholder="Introduce yourself or ask a question…"></textarea>
+        <div class="alert alert-info py-2 small mt-3">
+          🔒 By connecting, you share your name, age, gender, and conditions with the selected hospital.
+        </div>
+      </div>
+      <div class="modal-footer border-secondary">
+        <button class="btn btn-secondary btn-sm" data-bs-dismiss="modal">Cancel</button>
+        <button class="btn btn-primary btn-sm" id="sendConnBtn" disabled>Send Request</button>
+      </div>
+    </div>
+  </div>
+</div>
+
+<script src="https://cdn.jsdelivr.net/npm/bootstrap@5.3.2/dist/js/bootstrap.bundle.min.js"></script>
+<script>
+const PATIENT = {{ patient | tojson }};
+let conditions  = [...(PATIENT.conditions  || [])];
+let medications = [...(PATIENT.medications || [])];
+let documents   = [...(PATIENT.documents   || [])];
+let sysReady = false;
+let connectModal, msgModal, currentTrialId, currentTrialTitle, selectedHospId, currentConnId;
+
+// ── INIT ─────────────────────────────────────────────────────────────────────
+document.addEventListener('DOMContentLoaded', () => {
+  connectModal = new bootstrap.Modal(document.getElementById('connectModal'));
+  msgModal     = new bootstrap.Modal(document.getElementById('msgModal'));
+  document.getElementById('msgSendBtn').addEventListener('click', sendMsg);
+  document.getElementById('msgInput').addEventListener('keydown', e => {
+    if (e.key === 'Enter') sendMsg();
+  });
+  document.getElementById('docUploadBtn').addEventListener('click', uploadDoc);
+
+  document.querySelectorAll('#mainTabs .nav-link').forEach(b =>
+    b.addEventListener('click', () => switchTab(b.dataset.tab)));
+  document.getElementById('logoutBtn').addEventListener('click', logout);
+  document.getElementById('saveBtn').addEventListener('click', saveProfile);
+  document.getElementById('getMatchesBtn').addEventListener('click', loadTrials);
+  document.getElementById('refreshConnsBtn').addEventListener('click', loadConnections);
+  document.getElementById('sendConnBtn').addEventListener('click', submitConnect);
+
+  // Condition autocomplete
+  const ci = document.getElementById('condInput');
+  ci.addEventListener('input', () => condSearch(ci.value));
+  ci.addEventListener('keydown', e => { if(e.key==='Escape') hideCond(); });
+  document.addEventListener('click', e => {
+    if (!ci.contains(e.target) && !document.getElementById('condAutoList').contains(e.target))
+      hideCond();
+  });
+
+  // Medication input
+  document.getElementById('medInput').addEventListener('keydown', e => {
+    if (e.key === 'Enter') {
+      const v = e.target.value.trim();
+      if (v && !medications.includes(v)) { medications.push(v); renderMedTags(); }
+      e.target.value = '';
+    }
+  });
+
+  // Hospital selection in modal (event delegation)
+  document.getElementById('hospitalList').addEventListener('click', e => {
+    const item = e.target.closest('[data-hid]');
+    if (!item) return;
+    selectedHospId = item.dataset.hid;
+    document.querySelectorAll('#hospitalList [data-hid]')
+            .forEach(el => el.classList.remove('active'));
+    item.classList.add('active');
+    document.getElementById('sendConnBtn').disabled = false;
+  });
+
+  // Trial results event delegation (interest + connect)
+  document.getElementById('trialsResults').addEventListener('click', e => {
+    const btn = e.target.closest('[data-action]');
+    if (!btn) return;
+    const {action, trialId, trialTitle, score} = btn.dataset;
+    if (action === 'interest') toggleInterest(btn, trialId, trialTitle, parseFloat(score));
+    if (action === 'connect')  openConnectModal(trialId, trialTitle);
+  });
+
+  initProfile();
+  checkStatus().then(() => { if (!sysReady) startStatusPoll(); });
+});
+
+// ── STATUS ───────────────────────────────────────────────────────────────────
+let _statusPoll = null;
+async function checkStatus() {
+  try {
+    const d = await fetch('/api/status').then(r => r.json());
+    sysReady = !!d.ready;
+    document.getElementById('sysBanner').style.display = sysReady ? 'none' : '';
+    if (sysReady && _statusPoll) { clearInterval(_statusPoll); _statusPoll = null; }
+  } catch(e) {}
+}
+function startStatusPoll() {
+  if (_statusPoll) return;
+  _statusPoll = setInterval(async () => { await checkStatus(); }, 5000);
+}
+
+// ── TABS ─────────────────────────────────────────────────────────────────────
+function switchTab(name) {
+  document.querySelectorAll('#mainTabs .nav-link').forEach(b =>
+    b.classList.toggle('active', b.dataset.tab === name));
+  ['profile','trials','connections','inbox'].forEach(t =>
+    document.getElementById('tab-'+t).style.display = t === name ? '' : 'none');
+  if (name === 'connections') loadConnections();
+  if (name === 'inbox')       loadInbox();
+}
+
+// ── PROFILE ──────────────────────────────────────────────────────────────────
+function initProfile() {
+  document.getElementById('pf-fn').value       = PATIENT.first_name || '';
+  document.getElementById('pf-ln').value       = PATIENT.last_name  || '';
+  document.getElementById('pf-dob').value      = PATIENT.dob        || '';
+  document.getElementById('pf-gender').value   = PATIENT.gender     || '';
+  document.getElementById('pf-addr').value     = PATIENT.address    || '';
+  document.getElementById('openToTrials').checked = !!PATIENT.open_to_trials;
+  document.getElementById('acctInfo').innerHTML =
+    'Username: <strong>' + escH(PATIENT.username) + '</strong><br>' +
+    'Member since: <strong>' + escH((PATIENT.created_at||'').split('T')[0]) + '</strong>';
+  renderCondTags(); renderMedTags(); renderDocList();
+}
+
+async function saveProfile() {
+  const btn = document.getElementById('saveBtn');
+  const msg = document.getElementById('saveMsg');
+  btn.disabled = true; btn.textContent = 'Saving…'; msg.textContent = '';
+  try {
+    const r = await fetch('/api/patient/profile', {
+      method: 'POST', headers: {'Content-Type':'application/json'},
+      body: JSON.stringify({
+        first_name:    document.getElementById('pf-fn').value.trim(),
+        last_name:     document.getElementById('pf-ln').value.trim(),
+        dob:           document.getElementById('pf-dob').value,
+        gender:        document.getElementById('pf-gender').value,
+        address:       document.getElementById('pf-addr').value.trim(),
+        conditions, medications,
+        open_to_trials: document.getElementById('openToTrials').checked ? 1 : 0,
+      })
+    });
+    const d = await r.json();
+    msg.className = r.ok ? 'text-success small mt-2' : 'text-danger small mt-2';
+    msg.textContent = r.ok ? '✓ Profile saved' : (d.error || 'Save failed');
+  } catch(e) {
+    msg.className = 'text-danger small mt-2'; msg.textContent = 'Network error';
+  }
+  btn.disabled = false; btn.textContent = 'Save Profile';
+}
+
+// ── CONDITION TAGS (XSS-safe — no innerHTML for user data) ───────────────────
+function renderCondTags() {
+  const wrap = document.getElementById('condTags');
+  wrap.innerHTML = '';
+  conditions.forEach(v => wrap.appendChild(makeTag(v, 'bg-secondary', () => {
+    conditions = conditions.filter(c => c !== v); renderCondTags();
+  })));
+}
+
+function renderMedTags() {
+  const wrap = document.getElementById('medTags');
+  wrap.innerHTML = '';
+  medications.forEach(v => wrap.appendChild(makeTag(v, 'bg-info text-dark', () => {
+    medications = medications.filter(m => m !== v); renderMedTags();
+  })));
+}
+
+function makeTag(val, cls, onRemove) {
+  const span = document.createElement('span');
+  span.className = `badge ${cls} d-inline-flex align-items-center gap-1`;
+  const t = document.createElement('span');
+  t.textContent = val;
+  const x = document.createElement('button');
+  x.type = 'button'; x.className = 'btn-close btn-close-white'; x.style.fontSize = '.45rem';
+  if (cls.includes('text-dark')) x.className = 'btn-close';
+  x.addEventListener('click', onRemove);
+  span.appendChild(t); span.appendChild(x);
+  return span;
+}
+
+// ── CONDITION AUTOCOMPLETE ────────────────────────────────────────────────────
+async function condSearch(q) {
+  if (q.length < 2) { hideCond(); return; }
+  try {
+    const d = await fetch('/api/conditions/autocomplete?q=' + encodeURIComponent(q)).then(r=>r.json());
+    const list = document.getElementById('condAutoList');
+    list.innerHTML = '';
+    if (!(d.results||[]).length) { list.style.display='none'; return; }
+    d.results.forEach(item => {
+      const li = document.createElement('li');
+      li.textContent = item;
+      li.addEventListener('mousedown', e => {
+        e.preventDefault();
+        if (!conditions.includes(item)) { conditions.push(item); renderCondTags(); }
+        document.getElementById('condInput').value = ''; hideCond();
+      });
+      list.appendChild(li);
+    });
+    list.style.display = 'block';
+  } catch(e) { hideCond(); }
+}
+function hideCond() { document.getElementById('condAutoList').style.display = 'none'; }
+
+// ── TRIALS ───────────────────────────────────────────────────────────────────
+async function loadTrials() {
+  if (!sysReady) { await checkStatus(); }
+  if (!sysReady) { alert('Pipeline still loading. Please wait a moment.'); return; }
+  document.getElementById('trialsLoading').style.display = '';
+  document.getElementById('trialsMsg').style.display = 'none';
+  document.getElementById('trialsDisclaimer').style.display = 'none';
+  document.getElementById('trialsResults').innerHTML = '';
+  document.getElementById('getMatchesBtn').disabled = true;
+  try {
+    const r = await fetch('/api/patient/matches');
+    const d = await r.json();
+    if (!r.ok) throw new Error(d.error || 'Failed');
+    if (!(d.results||[]).length) {
+      const m = document.getElementById('trialsMsg');
+      m.style.display = ''; m.textContent = d.message || 'No matches found.';
+    } else {
+      renderTrials(d.results);
+    }
+  } catch(e) {
+    const m = document.getElementById('trialsMsg');
+    m.style.display = ''; m.className = 'text-danger text-center py-5';
+    m.textContent = e.message;
+  }
+  document.getElementById('trialsLoading').style.display = 'none';
+  document.getElementById('getMatchesBtn').disabled = false;
+}
+
+function renderTrials(trials) {
+  const wrap = document.getElementById('trialsResults');
+  wrap.innerHTML = '';
+  document.getElementById('trialsDisclaimer').style.display = trials.length ? '' : 'none';
+  trials.forEach(t => { wrap.insertAdjacentHTML('beforeend', buildCard(t)); });
+}
+
+function buildCard(t) {
+  const sc  = t.eligibility_probability >= 75 ? 'score-high' : t.eligibility_probability >= 50 ? 'score-mid' : 'score-low';
+  const stc = t.status === 'RECRUITING' ? 'success' : t.status.includes('ACTIVE') ? 'info' : 'secondary';
+  const overlaps = new Set(t.overlap_conditions || []);
+
+  const bars = [
+    ['Eligibility Probability', t.eligibility_probability,          'primary'],
+    ['Age Compatibility',       t.age_compatibility,                'info'],
+    ['Gender Compatibility',    t.gender_compatibility,             'success'],
+    ['Match Score',             t.match_score,                      'warning'],
+    ['Geo Feasibility',         t.geo_feasibility,                  'secondary'],
+    ['Med Compatibility',       t.med_compatibility,                'secondary'],
+    ['Rarity Score',            Math.round(t.condition_rarity_score*100), 'secondary'],
+  ].map(([l,v,c]) => `
+    <div class="mb-1">
+      <div class="d-flex justify-content-between">
+        <span class="bar-lbl">${escH(l)}</span>
+        <span class="bar-val">${v}%</span>
+      </div>
+      <div class="progress"><div class="progress-bar bg-${c}" style="width:${Math.min(100,v)}%"></div></div>
+    </div>`).join('');
+
+  const condTags = (t.trial_conditions||[]).map(c =>
+    `<span class="badge ${overlaps.has(c)?'bg-success':'bg-dark border border-secondary'} me-1 mb-1">${escH(c)}</span>`
+  ).join('');
+
+  const interested = t.interest_status === 'interested';
+
+  return `
+  <div class="trial-card mb-4">
+    <div class="d-flex gap-3 align-items-start mb-3">
+      <div class="score-circle ${sc}">
+        <span>${t.eligibility_probability}</span><small>%</small>
+      </div>
+      <div class="flex-grow-1 min-w-0">
+        <div class="text-muted small mb-1">${escH(t.trial_id)} &bull; ${escH(t.phase||'N/A')}</div>
+        <h6 class="mb-2">${escH(t.title)}</h6>
+        <div class="d-flex flex-wrap gap-1">
+          <span class="badge bg-${stc}">${escH(t.status)}</span>
+          ${t.phase&&t.phase!=='N/A'?`<span class="badge bg-info text-dark">${escH(t.phase)}</span>`:''}
+          <span class="badge bg-secondary">${escH(t.sex||'ALL')}</span>
+          <span class="badge bg-secondary">${t.min_age}–${t.max_age} yrs</span>
+        </div>
+      </div>
+    </div>
+    ${t.summary?`<p class="text-muted small mb-3">${escH(t.summary)}</p>`:''}
+    <div class="detail-grid mb-3">
+      <div><div class="lbl">Age Range</div><div class="val">${t.min_age}–${t.max_age} yrs</div></div>
+      <div><div class="lbl">Gender</div><div class="val">${escH(t.sex||'ALL')}</div></div>
+      <div><div class="lbl">Sites</div><div class="val">${t.n_sites||'—'}</div></div>
+      <div><div class="lbl">Enrollment</div><div class="val">${t.enrollment||'—'}</div></div>
+      ${t.facility_name?`<div class="col-span-2"><div class="lbl">Lead Site</div><div class="val">${escH(t.facility_name)}</div></div>`:''}
+      ${t.location?`<div class="col-span-2"><div class="lbl">Location</div><div class="val">${escH(t.location)}</div></div>`:''}
+    </div>
+    <div class="bar-section mb-3"><div class="sec-head">Match Analysis</div>${bars}</div>
+    <div class="mb-3">
+      <div class="sec-head">Conditions</div>
+      <div>${condTags}</div>
+      ${overlaps.size>0?'<div class="small text-success mt-1">🟢 Green = conditions matching your profile</div>':''}
+    </div>
+    ${t.criteria?`<details class="mb-3"><summary class="small text-muted" style="cursor:pointer">▶ Eligibility Criteria</summary>
+      <pre class="criteria-pre mt-2">${escH(t.criteria)}</pre></details>`:''}
+    <div class="d-flex flex-wrap gap-2 pt-2 border-top border-secondary mt-2">
+      <a href="https://clinicaltrials.gov/study/${escA(t.trial_id)}" target="_blank" rel="noopener"
+         class="btn btn-sm btn-outline-secondary">View on ClinicalTrials.gov ↗</a>
+      <button class="btn btn-sm ${interested?'btn-success':'btn-outline-success'}"
+              data-action="interest" data-trial-id="${escA(t.trial_id)}"
+              data-trial-title="${escA(t.title)}" data-score="${t.combined_score}">
+        ${interested?'✓ Interested':"I'm Interested"}</button>
+      <button class="btn btn-sm btn-outline-primary"
+              data-action="connect" data-trial-id="${escA(t.trial_id)}"
+              data-trial-title="${escA(t.title)}">Connect with Hospital</button>
+    </div>
+  </div>`;
+}
+
+async function toggleInterest(btn, trialId, title, score) {
+  const isOn = btn.classList.contains('btn-success');
+  if (isOn) {
+    await fetch('/api/patient/interest/'+encodeURIComponent(trialId), {method:'DELETE'});
+    btn.className = 'btn btn-sm btn-outline-success';
+    btn.textContent = "I'm Interested";
+  } else {
+    await fetch('/api/patient/interest', {
+      method:'POST', headers:{'Content-Type':'application/json'},
+      body: JSON.stringify({trial_id:trialId, trial_title:title, match_score:score})
+    });
+    btn.className = 'btn btn-sm btn-success';
+    btn.textContent = '✓ Interested';
+  }
+}
+
+// ── CONNECT WITH HOSPITAL ─────────────────────────────────────────────────────
+function _buildHospBtn(h) {
+  const btn = document.createElement('button');
+  btn.type = 'button';
+  btn.className = 'list-group-item list-group-item-action';
+  btn.dataset.hid = h.id;
+  const nm = document.createElement('strong'); nm.textContent = h.hospital_name;
+  const lc = document.createElement('small');
+  lc.className = 'd-block text-muted'; lc.textContent = h.location || '';
+  btn.appendChild(nm); btn.appendChild(lc);
+  if (h.match_reason) {
+    const badge = document.createElement('span');
+    badge.className = `badge ${h.match_tier === 1 ? 'bg-success' : 'bg-secondary'} ms-1`;
+    badge.style.fontSize = '.65rem';
+    badge.textContent = h.match_reason;
+    btn.appendChild(badge);
+  }
+  return btn;
+}
+
+async function openConnectModal(trialId, title) {
+  currentTrialId = trialId; currentTrialTitle = title;
+  selectedHospId = null;
+  document.getElementById('sendConnBtn').disabled = true;
+  document.getElementById('modalTrialInfo').textContent = 'Trial: ' + title;
+  document.getElementById('modalMsg').value = '';
+  const list = document.getElementById('hospitalList');
+  list.innerHTML = '<div class="list-group-item text-muted small">Loading hospitals…</div>';
+  connectModal.show();
+  try {
+    const d = await fetch('/api/patient/hospitals-for-trial?trial_id='+encodeURIComponent(trialId)).then(r=>r.json());
+    list.innerHTML = '';
+    const hospitals = d.hospitals || [];
+    if (!hospitals.length) {
+      list.innerHTML = '<div class="list-group-item text-muted small">No registered hospitals for this trial.</div>';
+      return;
+    }
+    const verified = hospitals.filter(h => h.match_tier === 1);
+    const related  = hospitals.filter(h => h.match_tier !== 1);
+
+    if (verified.length) {
+      const head = document.createElement('div');
+      head.className = 'sec-head text-success mb-1 mt-2';
+      head.textContent = 'Verified Trial Sites';
+      const grp = document.createElement('div');
+      grp.className = 'list-group mb-3';
+      verified.forEach(h => grp.appendChild(_buildHospBtn(h)));
+      list.appendChild(head); list.appendChild(grp);
+    }
+
+    if (related.length) {
+      const head = document.createElement('div');
+      head.className = 'sec-head mb-1 mt-2';
+      head.innerHTML = 'Related Hospitals'
+        + ' <span class="text-muted fw-normal" style="text-transform:none;letter-spacing:0;font-size:.68rem">'
+        + '— not confirmed trial sites</span>';
+      const grp = document.createElement('div');
+      grp.className = 'list-group mb-2';
+      related.forEach(h => grp.appendChild(_buildHospBtn(h)));
+      list.appendChild(head); list.appendChild(grp);
+    }
+  } catch(e) {
+    list.innerHTML = '<div class="list-group-item text-danger small">Error loading hospitals.</div>';
+  }
+}
+
+async function submitConnect() {
+  if (!selectedHospId) return;
+  try {
+    const r = await fetch('/api/patient/connect', {
+      method:'POST', headers:{'Content-Type':'application/json'},
+      body: JSON.stringify({hospital_id:selectedHospId, trial_id:currentTrialId,
+                            trial_title:currentTrialTitle,
+                            message:document.getElementById('modalMsg').value.trim()})
+    });
+    const d = await r.json();
+    if (r.ok) { connectModal.hide(); switchTab('connections'); }
+    else alert(d.error || 'Request failed');
+  } catch(e) { alert('Network error'); }
+}
+
+// ── CONNECTIONS ───────────────────────────────────────────────────────────────
+async function loadConnections() {
+  try {
+    const d = await fetch('/api/patient/connections').then(r=>r.json());
+    const conns = d.connections || [];
+    const msg  = document.getElementById('connsMsg');
+    const wrap = document.getElementById('connsWrap');
+    const tbody= document.getElementById('connsTbody');
+    if (!conns.length) { msg.style.display=''; wrap.style.display='none'; return; }
+    msg.style.display='none'; wrap.style.display='';
+    tbody.innerHTML = '';
+    conns.forEach(c => {
+      const sc  = {pending:'st-pending',accepted:'st-accepted',rejected:'st-rejected',completed:'st-completed'}[c.status]||'';
+      const tr  = document.createElement('tr');
+      const hname = c.hospital_name || '—';
+      tr.innerHTML = [
+        escH(hname),
+        escH((c.trial_title||'—').substring(0,50)),
+        escH(c.initiated_by||''),
+        `<span class="${sc} fw-bold">${escH(c.status)}</span>`,
+        escH((c.message||'').substring(0,60)),
+        escH((c.created_at||'').split('T')[0]),
+        `<button class="btn btn-sm btn-outline-info py-0 px-2" style="font-size:.72rem"
+           onclick="openMsgModal('${escA(c.id)}','${escA(hname)}')">💬 Chat</button>`,
+      ].map(v=>`<td>${v}</td>`).join('');
+      tbody.appendChild(tr);
+    });
+  } catch(e) {}
+}
+
+// ── DOCUMENTS ─────────────────────────────────────────────────────────────────
+function renderDocList() {
+  const wrap = document.getElementById('docList');
+  wrap.innerHTML = '';
+  if (!documents.length) {
+    const p = document.createElement('p');
+    p.className = 'text-muted small mb-1'; p.textContent = 'No documents uploaded yet.';
+    wrap.appendChild(p); return;
+  }
+  documents.forEach(doc => {
+    const row = document.createElement('div');
+    row.className = 'd-flex align-items-center gap-2 mb-1 small';
+    const nm  = document.createElement('span'); nm.className = 'flex-grow-1 text-truncate'; nm.textContent = doc.filename;
+    const dt  = document.createElement('small'); dt.className = 'text-muted flex-shrink-0';
+    dt.textContent = (doc.uploaded_at||'').substring(0,10);
+    const dl  = document.createElement('a');
+    dl.href = '/api/patient/documents/'+encodeURIComponent(doc.id)+'/download';
+    dl.className = 'btn btn-sm btn-outline-secondary py-0 px-1 flex-shrink-0';
+    dl.style.fontSize = '.7rem'; dl.textContent = '↓'; dl.title = 'Download';
+    const del = document.createElement('button');
+    del.type = 'button'; del.className = 'btn btn-sm btn-outline-danger py-0 px-1 flex-shrink-0';
+    del.style.fontSize = '.7rem'; del.textContent = '✕'; del.title = 'Delete';
+    del.addEventListener('click', () => deleteDoc(doc.id));
+    row.appendChild(nm); row.appendChild(dt); row.appendChild(dl); row.appendChild(del);
+    wrap.appendChild(row);
+  });
+}
+
+async function uploadDoc() {
+  const input = document.getElementById('docFileInput');
+  const msg   = document.getElementById('docMsg');
+  if (!input.files.length) { msg.className = 'text-warning small mt-1'; msg.textContent = 'Select a file first.'; return; }
+  const file = input.files[0];
+  if (file.size > 10 * 1024 * 1024) { msg.className = 'text-danger small mt-1'; msg.textContent = 'File too large (max 10 MB).'; return; }
+  const form = new FormData(); form.append('file', file);
+  msg.className = 'text-muted small mt-1'; msg.textContent = 'Uploading…';
+  try {
+    const r = await fetch('/api/patient/documents', {method:'POST', body:form});
+    const d = await r.json();
+    if (r.ok) {
+      documents.push(d.document); renderDocList(); input.value = '';
+      msg.className = 'text-success small mt-1'; msg.textContent = '✓ Uploaded';
+    } else {
+      msg.className = 'text-danger small mt-1'; msg.textContent = d.error || 'Upload failed';
+    }
+  } catch(e) { msg.className = 'text-danger small mt-1'; msg.textContent = 'Network error'; }
+}
+
+async function deleteDoc(docId) {
+  try {
+    const r = await fetch('/api/patient/documents/'+encodeURIComponent(docId), {method:'DELETE'});
+    if (r.ok) { documents = documents.filter(d => d.id !== docId); renderDocList(); }
+  } catch(e) {}
+}
+
+// ── MESSAGING ─────────────────────────────────────────────────────────────────
+async function openMsgModal(connId, label) {
+  currentConnId = connId;
+  document.getElementById('msgModalTitle').textContent = 'Chat — ' + label;
+  document.getElementById('msgThread').innerHTML =
+    '<div class="text-center text-muted small py-4">Loading…</div>';
+  document.getElementById('msgInput').value = '';
+  msgModal.show();
+  await loadMsgs();
+}
+
+async function loadMsgs() {
+  try {
+    const d = await fetch('/api/patient/connections/'+encodeURIComponent(currentConnId)+'/messages').then(r=>r.json());
+    renderMsgs(d.messages || []);
+  } catch(e) {
+    document.getElementById('msgThread').innerHTML =
+      '<div class="text-danger small text-center py-4">Error loading messages.</div>';
+  }
+}
+
+function renderMsgs(msgs) {
+  const thread = document.getElementById('msgThread');
+  if (!msgs.length) {
+    thread.innerHTML = '<div class="text-muted small text-center py-4">No messages yet. Start the conversation.</div>';
+    return;
+  }
+  thread.innerHTML = '';
+  msgs.forEach(m => {
+    const mine = m.sender_role === 'patient';
+    const wrap = document.createElement('div');
+    wrap.className = 'd-flex mb-2 ' + (mine ? 'justify-content-end' : 'justify-content-start');
+    const bubble = document.createElement('div');
+    bubble.style.cssText = 'max-width:72%;padding:.5rem .75rem;border-radius:12px;font-size:.85rem;'
+      + (mine ? 'background:#1f6feb;color:#fff;' : 'background:#21262d;color:#c9d1d9;');
+    const bodyEl = document.createElement('div'); bodyEl.textContent = m.body;
+    const ts = document.createElement('div');
+    ts.style.cssText = 'font-size:.65rem;opacity:.6;margin-top:.2rem;text-align:right';
+    ts.textContent = (m.created_at||'').replace('T',' ').substring(0,16);
+    bubble.appendChild(bodyEl); bubble.appendChild(ts);
+    wrap.appendChild(bubble); thread.appendChild(wrap);
+  });
+  thread.scrollTop = thread.scrollHeight;
+}
+
+async function sendMsg() {
+  const input = document.getElementById('msgInput');
+  const body  = input.value.trim();
+  if (!body || !currentConnId) return;
+  input.value = '';
+  try {
+    await fetch('/api/patient/connections/'+encodeURIComponent(currentConnId)+'/messages', {
+      method:'POST', headers:{'Content-Type':'application/json'},
+      body: JSON.stringify({body})
+    });
+    await loadMsgs();
+  } catch(e) {}
+}
+
+// ── INBOX ─────────────────────────────────────────────────────────────────────
+async function loadInbox() {
+  document.getElementById('inboxLoading').style.display = '';
+  document.getElementById('inboxEmpty').style.display   = 'none';
+  document.getElementById('inboxList').innerHTML = '';
+  try {
+    const d = await fetch('/api/patient/inbox').then(r => r.json());
+    document.getElementById('inboxLoading').style.display = 'none';
+    const threads = d.threads || [];
+    if (!threads.length) { document.getElementById('inboxEmpty').style.display = ''; return; }
+    let totalUnread = 0;
+    threads.forEach(t => {
+      totalUnread += t.unread_count || 0;
+      const hname   = t.hospital_name || 'Hospital';
+      const trial   = (t.trial_title || 'General Inquiry').substring(0, 60);
+      const preview = (t.last_message || 'No messages yet').substring(0, 100);
+      const card = document.createElement('div');
+      card.className = 'card bg-secondary mb-2';
+      const body = document.createElement('div');
+      body.className = 'card-body d-flex justify-content-between align-items-center gap-3';
+      const info = document.createElement('div');
+      info.style.minWidth = '0';
+      const nameEl = document.createElement('strong'); nameEl.textContent = hname;
+      info.appendChild(nameEl);
+      if (t.unread_count) {
+        const bEl = document.createElement('span');
+        bEl.className = 'badge bg-danger ms-2'; bEl.textContent = t.unread_count;
+        info.appendChild(bEl);
+      }
+      const trialEl = document.createElement('div');
+      trialEl.className = 'text-muted small text-truncate'; trialEl.textContent = trial;
+      const previewEl = document.createElement('div');
+      previewEl.className = 'text-secondary small fst-italic text-truncate'; previewEl.textContent = preview;
+      info.appendChild(trialEl); info.appendChild(previewEl);
+      const btn = document.createElement('button');
+      btn.className = 'btn btn-sm btn-outline-light flex-shrink-0';
+      btn.textContent = 'Open';
+      btn.onclick = () => openMsgModal(t.id, hname);
+      body.appendChild(info); body.appendChild(btn);
+      card.appendChild(body);
+      document.getElementById('inboxList').appendChild(card);
+    });
+    const badge = document.getElementById('inboxBadge');
+    if (totalUnread > 0) { badge.textContent = totalUnread; badge.style.display = ''; }
+    else badge.style.display = 'none';
+  } catch(e) {
+    document.getElementById('inboxLoading').style.display = 'none';
+    document.getElementById('inboxEmpty').style.display = '';
+  }
+}
+
+// ── UTILS ─────────────────────────────────────────────────────────────────────
+function escH(s) {
+  const d = document.createElement('div');
+  d.appendChild(document.createTextNode(String(s||'')));
+  return d.innerHTML;
+}
+function escA(s) {
+  return String(s||'').replace(/&/g,'&amp;').replace(/"/g,'&quot;').replace(/</g,'&lt;').replace(/>/g,'&gt;');
+}
+async function logout() {
+  await fetch('/auth/logout', {method:'POST'});
+  location.href = '/';
+}
+</script>
+</body>
+</html>