Upload app.py with huggingface_hub

app.py

@@ -28,12 +28,56 @@ KNOWHOW = ("MCL: Sylgard 184 PDMS 10:1 ratio 48hr cure green laser PIV 70bpm 5L/
     "TGT: Arduino Uno Stepper Motor 150mL blood sampled at 0 20 40 60 minutes. "
     "NORMAL RANGES: TAT below 8 ng/mL. PF1.2 below 2.0 nmol/L. Free hemoglobin below 20 mg/L. Platelets above 150 thousand per uL. "
     "HIGH RISK: TAT above 15. PF1.2 above 3.0. Hemoglobin above 50. Platelets below 100. "
-    "uPAD: Jaffe reaction creatinine picric acid orange-red. Normal creatinine 0.6-1.2 mg/dL. Borderline 1.2-1.5. CKD above 1.5. Stage2 1.5-3.0. Stage3-4 3.0-6.0. Stage5 above 6.0. "
-    "MHV: 27mm SJM Regent bileaflet also trileaflet monoleaflet pediatric designs. "
+    "uPAD: Jaffe reaction creatinine picric acid orange-red. Normal creatinine 0.6-1.2 mg/dL. Borderline 1.2-1.5. CKD above 1.5. "
+    "Stage2 1.5-3.0. Stage3-4 3.0-6.0. Stage5 above 6.0. "
+    "MHV: 27mm SJM Regent bileaflet also trileaflet monoleaflet pediatric. "
     "PIV: green laser 532nm time-resolved. Normal velocity 0.5-2.0 m/s. Normal shear below 5 Pa. Risk above 10 Pa. "
     "Equipment: Heska Element HT5 hematology analyzer time-resolved PIV Tygon tubing Arduino Uno stepper motor.")
 
-# ── LOAD PAPERS + FINE-TUNED MODEL ON STARTUP ─────────────────────
+CSS = """
+body, .gradio-container { background: #f7f7f8 !important; font-family: -apple-system, BlinkMacSystemFont, Segoe UI, sans-serif !important; }
+.tab-nav { background: #ffffff !important; border-bottom: 1px solid #e5e7eb !important; padding: 0 16px !important; display: flex !important; flex-wrap: wrap !important; }
+.tab-nav button { background: transparent !important; color: #6b7280 !important; border: none !important; border-bottom: 2px solid transparent !important; padding: 10px 12px !important; font-weight: 500 !important; font-size: 0.8em !important; white-space: nowrap !important; border-radius: 0 !important; }
+.tab-nav button:hover { color: #111827 !important; background: #f9fafb !important; }
+.tab-nav button.selected { color: #c1121f !important; border-bottom: 2px solid #c1121f !important; font-weight: 700 !important; background: transparent !important; }
+.message.user { background: #f3f4f6 !important; color: #1a202c !important; border-radius: 12px !important; }
+.message.bot { background: #ffffff !important; color: #1a202c !important; border-left: 3px solid #c1121f !important; }
+textarea { background: #ffffff !important; color: #1a202c !important; border: 1px solid #d1d5db !important; border-radius: 10px !important; }
+button.primary { background: #c1121f !important; color: white !important; border: none !important; border-radius: 8px !important; font-weight: 600 !important; }
+button.secondary { background: #f3f4f6 !important; color: #374151 !important; border: 1px solid #d1d5db !important; border-radius: 8px !important; }
+input[type=number] { background: #f9fafb !important; color: #1a202c !important; border: 1px solid #d1d5db !important; border-radius: 8px !important; }
+"""
+
+HEADER = """<div style="background:linear-gradient(135deg,#0a0f2e 0%,#1a0a0a 100%);padding:0;border-bottom:3px solid #c1121f;overflow:hidden;">
+<svg style="position:absolute;opacity:0.07;width:100%;height:100%;" viewBox="0 0 1200 120" preserveAspectRatio="none">
+<polyline points="0,60 100,60 130,20 150,100 170,10 200,90 220,60 400,60 430,20 450,100 470,10 500,90 520,60 700,60 730,20 750,100 770,10 800,90 820,60 1000,60 1030,20 1050,100 1070,10 1100,90 1120,60 1200,60" fill="none" stroke="#c1121f" stroke-width="3"/>
+</svg>
+<div style="max-width:1200px;margin:0 auto;padding:16px 24px;display:flex;align-items:center;justify-content:space-between;position:relative;z-index:1;">
+<div style="display:flex;align-items:center;gap:14px;">
+<svg width="55" height="55" viewBox="0 0 100 100"><circle cx="50" cy="35" r="28" fill="#0057a8" opacity="0.9"/><ellipse cx="50" cy="14" rx="22" ry="10" fill="#0057a8"/>
+<polygon points="30,14 33,4 36,14" fill="#e8a020"/><polygon points="36,12 39,2 42,12" fill="#e8a020"/>
+<polygon points="42,11 45,1 48,11" fill="#e8a020"/><polygon points="48,11 51,1 54,11" fill="#e8a020"/>
+<polygon points="54,12 57,2 60,12" fill="#e8a020"/><polygon points="60,14 63,4 66,14" fill="#e8a020"/>
+<rect x="36" y="30" width="28" height="22" rx="4" fill="#0057a8"/><rect x="40" y="35" width="8" height="12" rx="2" fill="#e8a020"/>
+<rect x="34" y="50" width="32" height="8" rx="4" fill="#0057a8"/></svg>
+<div><div style="color:#9ca3af;font-size:0.7em;letter-spacing:2px;text-transform:uppercase;">San Jose State University</div>
+<div style="color:#e8a020;font-size:0.82em;font-weight:700;">Biomedical Engineering</div></div></div>
+<div style="text-align:center;flex:1;padding:0 20px;">
+<div style="display:flex;align-items:center;justify-content:center;gap:10px;margin-bottom:3px;">
+<svg width="100" height="28" viewBox="0 0 120 32"><polyline points="0,16 20,16 26,4 30,28 34,2 38,26 44,16 120,16" fill="none" stroke="#c1121f" stroke-width="2.5" stroke-linecap="round"/></svg>
+<div style="font-size:2em;font-weight:900;letter-spacing:2px;"><span style="color:#ffffff;">Cardio</span><span style="color:#c1121f;">Lab</span><span style="color:#ffffff;"> AI</span></div>
+<svg width="100" height="28" viewBox="0 0 120 32" style="transform:scaleX(-1);"><polyline points="0,16 20,16 26,4 30,28 34,2 38,26 44,16 120,16" fill="none" stroke="#c1121f" stroke-width="2.5" stroke-linecap="round"/></svg></div>
+<div style="color:#9ca3af;font-size:0.68em;letter-spacing:2px;text-transform:uppercase;">RAG + Fine-tuned | Protocol Generator | Report Writer | BioGPT | 5 AI Models</div></div>
+<div style="display:flex;align-items:center;gap:14px;">
+<div style="text-align:right;"><div style="color:#9ca3af;font-size:0.68em;text-transform:uppercase;">Research Pillars</div>
+<div style="color:#ffffff;font-size:0.72em;margin-top:3px;">MHV  CKD  FSI</div>
+<div style="color:#9ca3af;font-size:0.62em;margin-top:2px;">MCL  PIV  TGT  uPAD  COMSOL</div></div>
+<svg width="48" height="48" viewBox="0 0 100 90">
+<path d="M50 85 C50 85 5 55 5 30 C5 15 18 5 30 5 C38 5 45 9 50 15 C55 9 62 5 70 5 C82 5 95 15 95 30 C95 55 50 85 50 85Z" fill="#c1121f" opacity="0.9"/>
+<polyline points="25,45 32,45 35,35 38,55 41,30 44,50 50,45 75,45" fill="none" stroke="white" stroke-width="2.5" stroke-linecap="round" opacity="0.9"/></svg></div></div>
+<div style="height:3px;background:linear-gradient(90deg,#0057a8,#c1121f,#e8a020,#c1121f,#0057a8);"></div></div>"""
+
+# ── PAPER DATABASE ─────────────────────────────────────────────────
 CHUNKS = []
 METADATA = []
 EMBEDDINGS = None
@@ -55,10 +99,10 @@ def load_papers():
         EMBEDDINGS = np.load(emb_path)
         EMBEDDER = SentenceTransformer("all-MiniLM-L6-v2")
         PAPERS_LOADED = True
-        print(f"Papers loaded: {len(CHUNKS)} chunks from {len(set(m['paper'] for m in METADATA))} papers")
+        print("Papers loaded: " + str(len(CHUNKS)) + " chunks")
         return True
     except Exception as e:
-        print(f"Paper load error: {e}")
+        print("Paper load error: " + str(e))
         return False
 
 def load_cardiolab_model():
@@ -66,29 +110,26 @@ def load_cardiolab_model():
     try:
         import torch
         from transformers import AutoModelForCausalLM, AutoTokenizer
-        from peft import PeftModel
         print("Loading CardioLab fine-tuned model...")
-        base_model = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
         CARDIOLAB_TOKENIZER = AutoTokenizer.from_pretrained(CARDIOLAB_MODEL, token=HF_TOKEN)
         CARDIOLAB_TOKENIZER.pad_token = CARDIOLAB_TOKENIZER.eos_token
         device = "cuda" if torch.cuda.is_available() else "cpu"
         CARDIOLAB_LLM = AutoModelForCausalLM.from_pretrained(
             CARDIOLAB_MODEL, token=HF_TOKEN,
-            torch_dtype=torch.float16 if device=="cuda" else torch.float32,
-            device_map="auto" if device=="cuda" else None,
+            torch_dtype=torch.float16 if device == "cuda" else torch.float32,
+            device_map="auto" if device == "cuda" else None,
             low_cpu_mem_usage=True
         )
         CARDIOLAB_MODEL_LOADED = True
-        print(f"CardioLab model loaded on {device}!")
+        print("CardioLab model loaded!")
         return True
     except Exception as e:
-        print(f"CardioLab model load error: {e}")
+        print("CardioLab model error: " + str(e))
         return False
 
 load_papers()
 load_cardiolab_model()
 
-# ── SEMANTIC SEARCH ────────────────────────────────────────────────
 def search_papers(query, n=4):
     if not PAPERS_LOADED or EMBEDDINGS is None or EMBEDDER is None:
         return "", []
@@ -107,85 +148,16 @@ def search_papers(query, n=4):
             meta = METADATA[idx]
             score = float(scores[idx])
             if score > 0.25:
-                results.append({"chunk":chunk,"paper":meta["paper"],"pillar":meta.get("pillar",""),"score":score})
+                results.append({"chunk": chunk, "paper": meta["paper"], "score": score})
                 if meta["paper"] not in seen:
-                    context += chr(10)+"=== FROM: "+meta["paper"]+" ==="+chr(10)
+                    context += chr(10) + "=== FROM: " + meta["paper"] + " ===" + chr(10)
                     seen.add(meta["paper"])
-                context += chunk[:500]+chr(10)
+                context += chunk[:500] + chr(10)
         return context, results
     except Exception as e:
         return "", []
 
-def answer_with_cardiolab_model(question, paper_context=""):
-    if not CARDIOLAB_MODEL_LOADED:
-        return None
-    try:
-        import torch
-        system = "You are CardioLab AI for SJSU Biomedical Engineering."
-        if paper_context:
-            system += " Use these SJSU research papers: "+paper_context[:500]
-        prompt = f"<|system|>{system}</s><|user|>{question}</s><|assistant|>"
-        inputs = CARDIOLAB_TOKENIZER(prompt, return_tensors="pt", truncation=True, max_length=512)
-        device = next(CARDIOLAB_LLM.parameters()).device
-        inputs = {k:v.to(device) for k,v in inputs.items()}
-        with torch.no_grad():
-            outputs = CARDIOLAB_LLM.generate(
-                **inputs, max_new_tokens=200, do_sample=True,
-                temperature=0.3, pad_token_id=CARDIOLAB_TOKENIZER.eos_token_id
-            )
-        response = CARDIOLAB_TOKENIZER.decode(outputs[0], skip_special_tokens=True)
-        if "<|assistant|>" in response:
-            answer = response.split("<|assistant|>")[-1].strip()
-        else:
-            answer = response[len(prompt):].strip() if len(response) > len(prompt) else response
-        return answer if len(answer) > 20 else None
-    except Exception as e:
-        print(f"CardioLab model error: {e}")
-        return None
-
-CSS = """
-body, .gradio-container { background: #f7f7f8 !important; font-family: -apple-system, BlinkMacSystemFont, Segoe UI, sans-serif !important; }
-.tab-nav { background: #ffffff !important; border-bottom: 1px solid #e5e7eb !important; padding: 0 16px !important; display: flex !important; flex-wrap: wrap !important; }
-.tab-nav button { background: transparent !important; color: #6b7280 !important; border: none !important; border-bottom: 2px solid transparent !important; padding: 10px 12px !important; font-weight: 500 !important; font-size: 0.8em !important; white-space: nowrap !important; border-radius: 0 !important; }
-.tab-nav button:hover { color: #111827 !important; background: #f9fafb !important; }
-.tab-nav button.selected { color: #c1121f !important; border-bottom: 2px solid #c1121f !important; font-weight: 700 !important; background: transparent !important; }
-.message.user { background: #f3f4f6 !important; color: #1a202c !important; border-radius: 12px !important; }
-.message.bot { background: #ffffff !important; color: #1a202c !important; border-left: 3px solid #c1121f !important; }
-textarea { background: #ffffff !important; color: #1a202c !important; border: 1px solid #d1d5db !important; border-radius: 10px !important; }
-button.primary { background: #c1121f !important; color: white !important; border: none !important; border-radius: 8px !important; font-weight: 600 !important; }
-button.secondary { background: #f3f4f6 !important; color: #374151 !important; border: 1px solid #d1d5db !important; border-radius: 8px !important; }
-input[type=number] { background: #f9fafb !important; color: #1a202c !important; border: 1px solid #d1d5db !important; border-radius: 8px !important; }
-"""
-
-HEADER = """<div style="background:linear-gradient(135deg,#0a0f2e 0%,#1a0a0a 100%);padding:0;border-bottom:3px solid #c1121f;overflow:hidden;">
-<svg style="position:absolute;opacity:0.07;width:100%;height:100%;" viewBox="0 0 1200 120" preserveAspectRatio="none">
-<polyline points="0,60 100,60 130,20 150,100 170,10 200,90 220,60 400,60 430,20 450,100 470,10 500,90 520,60 700,60 730,20 750,100 770,10 800,90 820,60 1000,60 1030,20 1050,100 1070,10 1100,90 1120,60 1200,60" fill="none" stroke="#c1121f" stroke-width="3"/>
-</svg>
-<div style="max-width:1200px;margin:0 auto;padding:16px 24px;display:flex;align-items:center;justify-content:space-between;position:relative;z-index:1;">
-<div style="display:flex;align-items:center;gap:14px;">
-<svg width="55" height="55" viewBox="0 0 100 100"><circle cx="50" cy="35" r="28" fill="#0057a8" opacity="0.9"/><ellipse cx="50" cy="14" rx="22" ry="10" fill="#0057a8"/>
-<polygon points="30,14 33,4 36,14" fill="#e8a020"/><polygon points="36,12 39,2 42,12" fill="#e8a020"/>
-<polygon points="42,11 45,1 48,11" fill="#e8a020"/><polygon points="48,11 51,1 54,11" fill="#e8a020"/>
-<polygon points="54,12 57,2 60,12" fill="#e8a020"/><polygon points="60,14 63,4 66,14" fill="#e8a020"/>
-<rect x="36" y="30" width="28" height="22" rx="4" fill="#0057a8"/><rect x="40" y="35" width="8" height="12" rx="2" fill="#e8a020"/>
-<rect x="34" y="50" width="32" height="8" rx="4" fill="#0057a8"/></svg>
-<div><div style="color:#9ca3af;font-size:0.7em;letter-spacing:2px;text-transform:uppercase;">San Jose State University</div>
-<div style="color:#e8a020;font-size:0.82em;font-weight:700;">Biomedical Engineering</div></div></div>
-<div style="text-align:center;flex:1;padding:0 20px;">
-<div style="display:flex;align-items:center;justify-content:center;gap:10px;margin-bottom:3px;">
-<svg width="100" height="28" viewBox="0 0 120 32"><polyline points="0,16 20,16 26,4 30,28 34,2 38,26 44,16 120,16" fill="none" stroke="#c1121f" stroke-width="2.5" stroke-linecap="round"/></svg>
-<div style="font-size:2em;font-weight:900;letter-spacing:2px;"><span style="color:#ffffff;">Cardio</span><span style="color:#c1121f;">Lab</span><span style="color:#ffffff;"> AI</span></div>
-<svg width="100" height="28" viewBox="0 0 120 32" style="transform:scaleX(-1);"><polyline points="0,16 20,16 26,4 30,28 34,2 38,26 44,16 120,16" fill="none" stroke="#c1121f" stroke-width="2.5" stroke-linecap="round"/></svg></div>
-<div style="color:#9ca3af;font-size:0.68em;letter-spacing:2px;text-transform:uppercase;">RAG + Fine-tuned | BioGPT | ClinicalTrials | Weekly Updates | 5 AI Models</div></div>
-<div style="display:flex;align-items:center;gap:14px;">
-<div style="text-align:right;"><div style="color:#9ca3af;font-size:0.68em;text-transform:uppercase;">Research Pillars</div>
-<div style="color:#ffffff;font-size:0.72em;margin-top:3px;">MHV  CKD  FSI</div>
-<div style="color:#9ca3af;font-size:0.62em;margin-top:2px;">MCL  PIV  TGT  uPAD  COMSOL</div></div>
-<svg width="48" height="48" viewBox="0 0 100 90">
-<path d="M50 85 C50 85 5 55 5 30 C5 15 18 5 30 5 C38 5 45 9 50 15 C55 9 62 5 70 5 C82 5 95 15 95 30 C95 55 50 85 50 85Z" fill="#c1121f" opacity="0.9"/>
-<polyline points="25,45 32,45 35,35 38,55 41,30 44,50 50,45 75,45" fill="none" stroke="white" stroke-width="2.5" stroke-linecap="round" opacity="0.9"/></svg></div></div>
-<div style="height:3px;background:linear-gradient(90deg,#0057a8,#c1121f,#e8a020,#c1121f,#0057a8);"></div></div>"""
-
+# ── SESSION MANAGEMENT ─────────────────────────────────────────────
 def load_all_sessions():
     if not HF_TOKEN: return {}
     try:
@@ -209,17 +181,17 @@ def get_session_list():
 
 def save_session(history, name):
     if not history: return "Nothing to save", gr.update()
-    if not name or not name.strip(): name = "Chat "+datetime.now().strftime("%b %d %H:%M")
+    if not name or not name.strip(): name = "Chat " + datetime.now().strftime("%b %d %H:%M")
     sessions = load_all_sessions()
-    sessions[name] = {"messages":history,"saved_at":datetime.now().isoformat()}
+    sessions[name] = {"messages": history, "saved_at": datetime.now().isoformat()}
     ok = save_all_sessions(sessions)
     choices = get_session_list()
-    return ("Saved: "+name if ok else "Save failed"), gr.update(choices=choices, value=name)
+    return ("Saved: " + name if ok else "Save failed"), gr.update(choices=choices, value=name)
 
 def load_session(name):
     if not name or "No saved" in name: return [], "Select a session"
     sessions = load_all_sessions()
-    return (sessions[name]["messages"], "Loaded: "+name) if name in sessions else ([], "Not found")
+    return (sessions[name]["messages"], "Loaded: " + name) if name in sessions else ([], "Not found")
 
 def delete_session(name):
     if not name or "No saved" in name: return "Select a session", gr.update()
@@ -227,11 +199,12 @@ def delete_session(name):
     if name in sessions:
         del sessions[name]; save_all_sessions(sessions)
         choices = get_session_list()
-        return "Deleted: "+name, gr.update(choices=choices, value=choices[0] if choices else None)
+        return "Deleted: " + name, gr.update(choices=choices, value=choices[0] if choices else None)
     return "Not found", gr.update()
 
 def new_chat(): return [], "", "New chat started"
 
+# ── SEARCH ─────────────────────────────────────────────────────────
 def get_pubmed_chat(query, n=3):
     try:
         r = requests.get("https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi",
@@ -240,190 +213,104 @@ def get_pubmed_chat(query, n=3):
         return chr(10).join(["https://pubmed.ncbi.nlm.nih.gov/"+i for i in ids]) if ids else ""
     except: return ""
 
-
-# ── PHASE C: BIOGPT + CLINICALTRIALS + WEEKLY UPDATE ──────────────
-
-def search_biogpt(query):
-    """Search BioGPT — trained on 15M PubMed papers via HuggingFace API"""
-    if not HF_TOKEN: return ""
+def expand_query_ai(query):
+    if not GROQ_KEY: return query
     try:
-        headers = {"Authorization": "Bearer "+HF_TOKEN}
-        # Use BioGPT for biomedical question answering
-        payload = {"inputs": query+" [SEP] Answer based on biomedical literature:"}
-        r = requests.post(
-            "https://api-inference.huggingface.co/models/microsoft/BioGPT-Large-PubMedQA",
-            headers=headers, json=payload, timeout=20
-        )
-        if r.status_code == 200:
-            result = r.json()
-            if isinstance(result, list) and len(result) > 0:
-                text = result[0].get("generated_text","")
-                # Extract just the answer part
-                if "[SEP]" in text:
-                    text = text.split("[SEP]")[-1].strip()
-                return text[:400] if text else ""
-        return ""
-    except: return ""
+        client = Groq(api_key=GROQ_KEY)
+        resp = client.chat.completions.create(model="llama-3.1-8b-instant",
+            messages=[{"role":"system","content":"Biomedical PubMed expert. Convert to MeSH terms for heart valves hemodynamics PIV thrombogenicity FSI microfluidics CKD. Return ONLY terms."},
+            {"role":"user","content":"Optimize: " + query}], max_tokens=80)
+        return resp.choices[0].message.content.strip() or query
+    except: return query
 
-def search_clinical_trials(query, n=5):
-    """Search ClinicalTrials.gov for heart valve and CKD trials"""
+def quick_search(query, search_model="Llama 3.3 70B (Best)"):
+    if not query.strip(): return "Please enter a topic."
+    expanded = expand_query_ai(query)
+    results = []
     try:
-        r = requests.get(
-            "https://clinicaltrials.gov/api/v2/studies",
-            params={
-                "query.term": query,
-                "filter.overallStatus": "RECRUITING|COMPLETED",
-                "pageSize": n,
-                "format": "json",
-                "fields": "NCTId,BriefTitle,OverallStatus,Phase,StartDate,Condition"
-            },
-            timeout=12
-        )
-        if r.status_code != 200: return []
-        studies = r.json().get("studies",[])
-        results = []
-        for s in studies:
-            proto = s.get("protocolSection",{})
-            ident = proto.get("identificationModule",{})
-            status = proto.get("statusModule",{})
-            nct = ident.get("nctId","")
-            title = ident.get("briefTitle","")
-            phase = status.get("phase","")
-            overall = status.get("overallStatus","")
-            if nct and title:
-                results.append({
-                    "nct": nct,
-                    "title": title,
-                    "status": overall,
-                    "phase": phase,
-                    "url": "https://clinicaltrials.gov/study/"+nct
-                })
-        return results
-    except: return []
-
-def get_weekly_pubmed_update(topics=None):
-    """Get papers published in last 7 days on CardioLab topics"""
-    if topics is None:
-        topics = [
-            "mechanical heart valve thrombogenicity",
-            "microfluidic creatinine CKD diagnosis",
-            "PIV hemodynamics prosthetic valve",
-            "Mock Circulatory Loop cardiac",
-            "bileaflet valve fluid structure interaction"
-        ]
-    all_new = []
+        forced = expanded + " AND (heart valve OR hemodynamics OR microfluidic OR thrombogen OR creatinine OR PIV OR CFD OR CKD)"
+        r = requests.get("https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi",
+            params={"db":"pubmed","term":forced,"retmax":8,"retmode":"json","sort":"date","field":"tiab"},timeout=12)
+        ids = r.json()["esearchresult"]["idlist"]
+        if ids:
+            r2 = requests.get("https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi",
+                params={"db":"pubmed","id":",".join(ids),"retmode":"xml","rettype":"abstract"},timeout=12)
+            import xml.etree.ElementTree as ET
+            root = ET.fromstring(r2.content)
+            for article in root.findall(".//PubmedArticle"):
+                try:
+                    title = article.find(".//ArticleTitle").text or "No title"
+                    pmid = article.find(".//PMID").text or ""
+                    year_el = article.find(".//PubDate/Year")
+                    year = year_el.text if year_el is not None else ""
+                    results.append({"source":"PubMed","title":str(title),"year":year,"url":"https://pubmed.ncbi.nlm.nih.gov/"+pmid})
+                except: continue
+    except: pass
     try:
-        from datetime import datetime, timedelta
-        week_ago = (datetime.now() - timedelta(days=7)).strftime("%Y/%m/%d")
-        for topic in topics:
-            r = requests.get(
-                "https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi",
-                params={
-                    "db":"pubmed",
-                    "term":topic,
-                    "mindate":week_ago,
-                    "datetype":"pdat",
-                    "retmax":3,
-                    "retmode":"json",
-                    "sort":"date"
-                },
-                timeout=10
-            )
-            ids = r.json()["esearchresult"]["idlist"]
-            for pmid in ids:
-                all_new.append({
-                    "pmid": pmid,
-                    "topic": topic,
-                    "url": "https://pubmed.ncbi.nlm.nih.gov/"+pmid
-                })
-        return all_new
-    except: return []
-
-def full_research_search(query, search_model="Llama 3.3 70B (Best)"):
-    """Complete search across ALL sources including Phase C additions"""
-    if not query.strip(): return "Please enter a research topic."
-
-    model_id = CHAT_MODELS.get(search_model, "llama-3.3-70b-versatile")
-    expanded = expand_query_ai(query, model_id) if GROQ_KEY else query
-
-    # All search sources
-    pubmed = fetch_pubmed(expanded, n=6)
-    scholar = fetch_scholar(expanded, n=5)
-    europe = fetch_europe_pmc(expanded, n=4)
-    trials = search_clinical_trials(query, n=4)
-    weekly = get_weekly_pubmed_update()
-    biogpt_answer = search_biogpt(query)
-
-    # Format output
-    out = "QUERY: "+query+chr(10)
-    out += "AI EXPANDED: "+expanded+chr(10)
-    out += "SOURCES: PubMed + Scholar + EuropePMC + ClinicalTrials + SJSU + BioGPT"+chr(10)
-    out += "="*50+chr(10)+chr(10)
-
-    # BioGPT answer first
-    if biogpt_answer:
-        out += "BIOGPT ANSWER (trained on 15M PubMed papers):"+chr(10)
-        out += biogpt_answer+chr(10)+chr(10)
-        out += "="*50+chr(10)+chr(10)
-
-    # PubMed results
-    if pubmed:
-        out += "PUBMED ("+str(len(pubmed))+" papers):"+chr(10)
-        for p in pubmed[:6]:
-            out += p["title"][:85]+" ("+p["year"]+")"+chr(10)
-            out += "  "+p["url"]+chr(10)+chr(10)
-
-    # Scholar results
-    if scholar:
-        out += "SEMANTIC SCHOLAR ("+str(len(scholar))+" papers):"+chr(10)
-        for p in scholar[:5]:
-            out += p["title"][:85]+" ("+p["year"]+")"
-            if p["citations"] not in ("N/A","","0"): out += " | "+p["citations"]+" citations"
-            out += chr(10)+"  "+p["url"]+chr(10)+chr(10)
-
-    # Clinical trials
-    if trials:
-        out += "CLINICALTRIALS.GOV ("+str(len(trials))+" trials):"+chr(10)
-        for t in trials:
-            out += t["title"][:80]+" | "+t["status"]+" | "+t.get("phase","")+" "+chr(10)
-            out += "  "+t["url"]+chr(10)+chr(10)
-
-    # Weekly updates
-    weekly_relevant = [w for w in weekly if any(
-        kw in query.lower() for kw in ["valve","heart","ckd","creatinine","piv","tgt","mcl"]
-    )]
-    if weekly_relevant:
-        out += "NEW THIS WEEK (last 7 days):"+chr(10)
-        for w in weekly_relevant[:5]:
-            out += "  "+w["url"]+" ["+w["topic"][:40]+"]"+chr(10)
-
-    # SJSU ScholarWorks
-    out += chr(10)+"SJSU SCHOLARWORKS:"+chr(10)
-    out += "  https://scholarworks.sjsu.edu/do/search/?q="+requests.utils.quote(query)+"&context=6781027"
-
+        r = requests.get("https://api.semanticscholar.org/graph/v1/paper/search",
+            params={"query":expanded,"limit":6,"fields":"title,year,url,citationCount"},timeout=12)
+        for p in r.json().get("data",[]):
+            year = p.get("year",0) or 0
+            if int(year) >= 2015:
+                results.append({"source":"Scholar","title":p.get("title",""),"year":str(year),"url":p.get("url",""),"citations":str(p.get("citationCount",0))})
+    except: pass
+    out = "QUERY: " + query + chr(10) + "AI EXPANDED: " + expanded + chr(10) + "="*45 + chr(10) + chr(10)
+    groups = {"PubMed":[],"Scholar":[]}
+    seen = set()
+    for r in results:
+        key = r["title"][:50].lower()
+        if key not in seen and r["url"]:
+            seen.add(key); groups[r["source"]].append(r)
+    for source, papers in groups.items():
+        if not papers: continue
+        out += "--- " + source + " ---" + chr(10)
+        for p in papers[:8]:
+            out += p["title"][:85] + " (" + p["year"] + ")" + chr(10)
+            out += "  " + p["url"] + chr(10) + chr(10)
+    out += "--- SJSU ScholarWorks ---" + chr(10)
+    out += "https://scholarworks.sjsu.edu/do/search/?q=" + requests.utils.quote(query) + "&context=6781027"
     return out
 
+# ── CHAT ───────────────────────────────────────────────────────────
+def answer_with_cardiolab_model(question, paper_context=""):
+    if not CARDIOLAB_MODEL_LOADED: return None
+    try:
+        import torch
+        system = "You are CardioLab AI for SJSU Biomedical Engineering."
+        if paper_context: system += " Use these SJSU research papers: " + paper_context[:400]
+        prompt = "<|system|>" + system + "</s><|user|>" + question + "</s><|assistant|>"
+        inputs = CARDIOLAB_TOKENIZER(prompt, return_tensors="pt", truncation=True, max_length=512)
+        device = next(CARDIOLAB_LLM.parameters()).device
+        inputs = {k: v.to(device) for k, v in inputs.items()}
+        with torch.no_grad():
+            outputs = CARDIOLAB_LLM.generate(**inputs, max_new_tokens=200, do_sample=True,
+                temperature=0.3, pad_token_id=CARDIOLAB_TOKENIZER.eos_token_id)
+        response = CARDIOLAB_TOKENIZER.decode(outputs[0], skip_special_tokens=True)
+        if "<|assistant|>" in response:
+            answer = response.split("<|assistant|>")[-1].strip()
+        else:
+            answer = response[-300:].strip()
+        return answer if len(answer) > 20 else None
+    except Exception as e:
+        print("CardioLab model error: " + str(e))
+        return None
 
 def research_chat(message, history, chat_model="Llama 3.3 70B (Best)"):
     if not message.strip(): return "", history
     paper_context, paper_results = search_papers(message, n=4)
-
-    # Use fine-tuned CardioLab model if selected
     if chat_model == "CardioLab Fine-tuned (SJSU)" and CARDIOLAB_MODEL_LOADED:
         answer = answer_with_cardiolab_model(message, paper_context)
         if answer:
             if paper_results:
                 unique_papers = list(dict.fromkeys([r["paper"] for r in paper_results]))
-                answer += chr(10)+chr(10)+"Sources from SJSU CardioLab papers:"
+                answer += chr(10) + chr(10) + "Sources from SJSU CardioLab papers:"
                 for p in unique_papers[:3]:
-                    answer += chr(10)+"  - "+p.replace('.pdf','').replace('_',' ')
+                    answer += chr(10) + "  - " + p.replace(".pdf","").replace("_"," ")
             pubmed = get_pubmed_chat(message, n=2)
-            if pubmed: answer += chr(10)+"PubMed: "+pubmed
+            if pubmed: answer += chr(10) + "PubMed: " + pubmed
             history.append({"role":"user","content":message})
-            history.append({"role":"assistant","content":"[CardioLab Fine-tuned Model] "+answer})
+            history.append({"role":"assistant","content":"[CardioLab Fine-tuned] " + answer})
             return "", history
-
-    # Fall back to Groq models
     if not GROQ_KEY:
         history.append({"role":"user","content":message})
         history.append({"role":"assistant","content":"Error: Add GROQ_API_KEY to Space Settings."})
@@ -433,12 +320,11 @@ def research_chat(message, history, chat_model="Llama 3.3 70B (Best)"):
         client = Groq(api_key=GROQ_KEY)
         if paper_context:
             system_prompt = ("You are CardioLab AI for SJSU Biomedical Engineering. "
-                "Answer using SJSU CardioLab research papers below. "
-                "Always cite the paper name when using specific data."+chr(10)+chr(10)+
-                "SJSU CARDIOLAB PAPERS:"+chr(10)+paper_context+chr(10)+chr(10)+
-                "ADDITIONAL KNOWLEDGE: "+KNOWHOW)
+                "Answer using SJSU CardioLab research papers below. Cite paper names with specific data." +
+                chr(10) + chr(10) + "SJSU CARDIOLAB PAPERS:" + chr(10) + paper_context +
+                chr(10) + chr(10) + "ADDITIONAL KNOWLEDGE: " + KNOWHOW)
         else:
-            system_prompt = "You are CardioLab AI for SJSU Biomedical Engineering. Expert in MHV MCL PIV TGT uPAD CKD FSI. "+KNOWHOW
+            system_prompt = "You are CardioLab AI for SJSU Biomedical Engineering. Expert in MHV MCL PIV TGT uPAD CKD FSI. " + KNOWHOW
         msgs = [{"role":"system","content":system_prompt}]
         for item in history:
             if isinstance(item, dict): msgs.append({"role":item["role"],"content":item["content"]})
@@ -447,17 +333,17 @@ def research_chat(message, history, chat_model="Llama 3.3 70B (Best)"):
         answer = resp.choices[0].message.content
         if paper_results:
             unique_papers = list(dict.fromkeys([r["paper"] for r in paper_results]))
-            answer += chr(10)+chr(10)+"Sources from SJSU CardioLab papers:"
+            answer += chr(10) + chr(10) + "Sources from SJSU CardioLab papers:"
             for p in unique_papers[:3]:
-                answer += chr(10)+"  - "+p.replace('.pdf','').replace('_',' ')
+                answer += chr(10) + "  - " + p.replace(".pdf","").replace("_"," ")
         pubmed = get_pubmed_chat(message, n=2)
-        if pubmed: answer += chr(10)+"PubMed: "+pubmed
+        if pubmed: answer += chr(10) + "PubMed: " + pubmed
         history.append({"role":"user","content":message})
         history.append({"role":"assistant","content":answer})
         return "", history
     except Exception as e:
         history.append({"role":"user","content":message})
-        history.append({"role":"assistant","content":"Error: "+str(e)})
+        history.append({"role":"assistant","content":"Error: " + str(e)})
         return "", history
 
 def voice_chat(audio, history):
@@ -469,79 +355,120 @@ def voice_chat(audio, history):
         with open(audio, "rb") as f:
             tx = client.audio.transcriptions.create(file=("audio.wav", f, "audio/wav"), model="whisper-large-v3")
         paper_context, _ = search_papers(tx.text, n=3)
-        system = "You are CardioLab AI. "+KNOWHOW
-        if paper_context: system = "You are CardioLab AI. Use these SJSU papers:"+chr(10)+paper_context+chr(10)+KNOWHOW
+        system = "You are CardioLab AI. " + KNOWHOW
+        if paper_context: system = "You are CardioLab AI. Use these SJSU papers:" + chr(10) + paper_context + chr(10) + KNOWHOW
         msgs = [{"role":"system","content":system}]
         for item in history:
             if isinstance(item, dict): msgs.append({"role":item["role"],"content":item["content"]})
         msgs.append({"role":"user","content":tx.text})
-        resp = client.chat.completions.create(model="llama-3.3-70b-versatile",messages=msgs,max_tokens=500)
-        history.append({"role":"user","content":"Voice: "+tx.text})
+        resp = client.chat.completions.create(model="llama-3.3-70b-versatile", messages=msgs, max_tokens=500)
+        history.append({"role":"user","content":"Voice: " + tx.text})
         history.append({"role":"assistant","content":resp.choices[0].message.content})
         return history
     except Exception as e:
-        history.append({"role":"assistant","content":"Voice error: "+str(e)})
+        history.append({"role":"assistant","content":"Voice error: " + str(e)})
         return history
 
-def expand_query_ai(query):
-    if not GROQ_KEY: return query
+# ── PHASE D: PROTOCOL GENERATOR + REPORT WRITER + HYPOTHESIS ──────
+def generate_protocol(experiment_type, specific_params):
+    if not GROQ_KEY: return "Error: Add GROQ_API_KEY to Space Settings."
+    if not experiment_type: return "Please select an experiment type."
     try:
         client = Groq(api_key=GROQ_KEY)
-        resp = client.chat.completions.create(model="llama-3.1-8b-instant",
-            messages=[{"role":"system","content":"Biomedical PubMed expert. Convert to MeSH terms for heart valves hemodynamics PIV thrombogenicity FSI microfluidics CKD. Return ONLY terms."},
-            {"role":"user","content":"Optimize: "+query}],max_tokens=80)
-        return resp.choices[0].message.content.strip() or query
-    except: return query
+        paper_context, _ = search_papers(experiment_type, n=4)
+        lab_context = {
+            "MCL": "Sylgard 184 PDMS 10:1 ratio 48hr cure. Tygon tubing. 70bpm 5L/min 80-120mmHg.",
+            "PIV": "Green laser 532nm time-resolved. Normal velocity 0.5-2.0 m/s. Shear below 5 Pa.",
+            "TGT": "Arduino Uno stepper motor 48V. 150mL fresh blood. Sample at 0 20 40 60 min. Heska HT5.",
+            "uPAD": "Whatman filter paper. Wax printer 120C. Picric acid alkaline solution. Jaffe reaction.",
+            "FSI": "COMSOL Multiphysics ALE mesh. Blood 1060 kg/m3 0.0035 Pa.s. SJM bileaflet geometry.",
+        }
+        extra = next((v for k, v in lab_context.items() if k.lower() in experiment_type.lower()), "")
+        system_msg = ("You are CardioLab AI protocol generator for SJSU Biomedical Engineering. "
+            "Generate a COMPLETE detailed lab protocol with these sections: "
+            "1. OBJECTIVE "
+            "2. MATERIALS AND EQUIPMENT with exact quantities "
+            "3. SAFETY CONSIDERATIONS "
+            "4. STEP-BY-STEP PROCEDURE numbered and detailed "
+            "5. DATA COLLECTION "
+            "6. ANALYSIS METHOD "
+            "7. EXPECTED RESULTS with normal ranges "
+            "8. TROUBLESHOOTING "
+            "Use exact SJSU CardioLab values and equipment.")
+        user_msg = "Generate complete protocol for: " + experiment_type
+        if specific_params and specific_params.strip():
+            user_msg += chr(10) + "Parameters: " + specific_params
+        if extra:
+            user_msg += chr(10) + "CardioLab context: " + extra
+        if paper_context:
+            user_msg += chr(10) + "From SJSU papers: " + paper_context[:600]
+        resp = client.chat.completions.create(
+            model="llama-3.3-70b-versatile",
+            messages=[{"role":"system","content":system_msg},{"role":"user","content":user_msg}],
+            max_tokens=1200)
+        return resp.choices[0].message.content
+    except Exception as e:
+        return "Error generating protocol: " + str(e)
 
-def quick_search(query, search_model="Llama 3.3 70B (Best)"):
-    if not query.strip(): return "Please enter a topic."
-    expanded = expand_query_ai(query)
-    results = []
+def generate_report(data_description, experiment_type, results):
+    if not GROQ_KEY: return "Error: Add GROQ_API_KEY to Space Settings."
+    if not experiment_type: return "Please select a study type."
     try:
-        forced = expanded+" AND (heart valve OR hemodynamics OR microfluidic OR thrombogen OR creatinine OR PIV OR CFD OR CKD)"
-        r = requests.get("https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi",
-            params={"db":"pubmed","term":forced,"retmax":8,"retmode":"json","sort":"date","field":"tiab"},timeout=12)
-        ids = r.json()["esearchresult"]["idlist"]
-        if ids:
-            r2 = requests.get("https://eutils.ncbi.nlm.nih.gov/entrez/eutils/efetch.fcgi",
-                params={"db":"pubmed","id":",".join(ids),"retmode":"xml","rettype":"abstract"},timeout=12)
-            import xml.etree.ElementTree as ET
-            root = ET.fromstring(r2.content)
-            for article in root.findall(".//PubmedArticle"):
-                try:
-                    title = article.find(".//ArticleTitle").text or "No title"
-                    pmid = article.find(".//PMID").text or ""
-                    year_el = article.find(".//PubDate/Year")
-                    year = year_el.text if year_el is not None else ""
-                    results.append({"source":"PubMed","title":str(title),"year":year,"url":"https://pubmed.ncbi.nlm.nih.gov/"+pmid,"citations":"N/A"})
-                except: continue
-    except: pass
+        client = Groq(api_key=GROQ_KEY)
+        paper_context, _ = search_papers(experiment_type, n=3)
+        system_msg = ("You are CardioLab AI report writer for SJSU Biomedical Engineering. "
+            "Generate a professional research report with these sections: "
+            "1. ABSTRACT 150 words "
+            "2. INTRODUCTION background and objectives "
+            "3. MATERIALS AND METHODS "
+            "4. RESULTS AND DISCUSSION "
+            "5. CONCLUSION "
+            "6. RECOMMENDATIONS "
+            "7. REFERENCES cite SJSU CardioLab papers "
+            "Use specific values. Write in professional academic style.")
+        user_msg = "Write research report for: " + experiment_type
+        if data_description and data_description.strip():
+            user_msg += chr(10) + "Description: " + data_description
+        if results and results.strip():
+            user_msg += chr(10) + "Results: " + results
+        if paper_context:
+            user_msg += chr(10) + "SJSU papers: " + paper_context[:600]
+        resp = client.chat.completions.create(
+            model="llama-3.3-70b-versatile",
+            messages=[{"role":"system","content":system_msg},{"role":"user","content":user_msg}],
+            max_tokens=1500)
+        return resp.choices[0].message.content
+    except Exception as e:
+        return "Error generating report: " + str(e)
+
+def generate_hypothesis(research_area, current_findings):
+    if not GROQ_KEY: return "Error: Add GROQ_API_KEY to Space Settings."
+    if not research_area: return "Please select a research area."
     try:
-        r = requests.get("https://api.semanticscholar.org/graph/v1/paper/search",
-            params={"query":expanded,"limit":6,"fields":"title,year,url,citationCount"},timeout=12)
-        for p in r.json().get("data",[]):
-            year = p.get("year",0) or 0
-            if int(year) >= 2015:
-                results.append({"source":"Scholar","title":p.get("title",""),"year":str(year),"url":p.get("url",""),"citations":str(p.get("citationCount",0))})
-    except: pass
-    out = "QUERY: "+query+chr(10)+"AI EXPANDED: "+expanded+chr(10)+"="*45+chr(10)+chr(10)
-    groups = {"PubMed":[],"Scholar":[]}
-    seen = set()
-    for r in results:
-        key = r["title"][:50].lower()
-        if key not in seen and r["url"]:
-            seen.add(key); groups[r["source"]].append(r)
-    for source, papers in groups.items():
-        if not papers: continue
-        out += "--- "+source+" ---"+chr(10)
-        for p in papers[:8]:
-            out += p["title"][:85]+" ("+p["year"]+")"
-            if p["citations"] not in ("N/A","","0"): out += " | "+p["citations"]+" citations"
-            out += chr(10)+"  "+p["url"]+chr(10)+chr(10)
-    out += "--- SJSU ScholarWorks ---"+chr(10)
-    out += "https://scholarworks.sjsu.edu/do/search/?q="+requests.utils.quote(query)+"&context=6781027"
-    return out
+        client = Groq(api_key=GROQ_KEY)
+        paper_context, _ = search_papers(research_area, n=3)
+        system_msg = ("You are CardioLab AI research assistant for SJSU Biomedical Engineering. "
+            "Generate 3 specific testable research hypotheses. For each provide: "
+            "H0 null hypothesis, "
+            "H1 alternative hypothesis, "
+            "Scientific rationale, "
+            "Suggested experiment, "
+            "Expected outcome and measurable metrics. "
+            "Base on SJSU CardioLab research.")
+        user_msg = "Generate hypotheses for: " + research_area
+        if current_findings and current_findings.strip():
+            user_msg += chr(10) + "Current findings: " + current_findings
+        if paper_context:
+            user_msg += chr(10) + "SJSU papers: " + paper_context[:500]
+        resp = client.chat.completions.create(
+            model="llama-3.3-70b-versatile",
+            messages=[{"role":"system","content":system_msg},{"role":"user","content":user_msg}],
+            max_tokens=1000)
+        return resp.choices[0].message.content
+    except Exception as e:
+        return "Error: " + str(e)
 
+# ── ANALYSIS TOOLS ─────────────────────────────────────────────────
 def analyze_upad_photo(image):
     if image is None: return None, "Upload a uPAD photo first."
     try:
@@ -587,7 +514,7 @@ def analyze_piv_csv(file,theme="White"):
             if vc:
                 ax.plot(xv,df[vc],color="#c1121f",linewidth=2.5,marker="o",markersize=5)
                 ax.fill_between(xv,df[vc],alpha=0.15,color="#c1121f")
-                ax.axhline(y=2.0,color="#f59e0b",linestyle="--",linewidth=2,label="Risk: 2.0 m/s")
+                ax.axhline(y=2.0,color="#f59e0b",linestyle="--",linewidth=2,label="Risk 2.0 m/s")
                 ax.set_ylabel("Velocity (m/s)",color=ac); ax.legend(fontsize=9,labelcolor=fg,facecolor=pb)
         def ps(ax):
             if sc2:
@@ -595,7 +522,7 @@ def analyze_piv_csv(file,theme="White"):
                 ax.plot(xp,df[sc2],color="#0057a8",linewidth=2.5,marker="s",markersize=5)
                 ax.fill_between(xp,df[sc2],alpha=0.15,color="#0057a8")
                 ax.axhline(y=5,color="#f59e0b",linestyle="--",linewidth=2,label="Caution 5 Pa")
-                ax.axhline(y=10,color="#c1121f",linestyle="--",linewidth=2,label="High risk 10 Pa")
+                ax.axhline(y=10,color="#c1121f",linestyle="--",linewidth=2,label="Risk 10 Pa")
                 ax.set_ylabel("Shear (Pa)",color=ac); ax.legend(fontsize=9,labelcolor=fg,facecolor=pb)
         def psc(ax):
             if vc and sc2:
@@ -665,7 +592,7 @@ def analyze_tgt_csv(file,theme="White"):
             try:
                 client=Groq(api_key=GROQ_KEY)
                 resp=client.chat.completions.create(model="llama-3.3-70b-versatile",
-                    messages=[{"role":"system","content":"Hematology expert. Give thrombogenicity risk."},
+                    messages=[{"role":"system","content":"Hematology expert. Thrombogenicity risk."},
                     {"role":"user","content":"TGT:"+chr(10)+df.describe().to_string()[:500]}],max_tokens=250)
                 ai=chr(10)+"AI: "+resp.choices[0].message.content
             except: pass
@@ -701,118 +628,27 @@ def generate_image(prompt):
 def piv_manual(v,s,h):
     vr="HIGH-stenosis" if float(v)>2.0 else "NORMAL"
     sr="HIGH-thrombosis" if float(s)>10 else "ELEVATED" if float(s)>5 else "NORMAL"
-    return "Velocity: "+str(v)+" m/s — "+vr+chr(10)+"Shear: "+str(s)+" Pa — "+sr+chr(10)+"HR: "+str(h)+" bpm"
+    return "Velocity: "+str(v)+" m/s - "+vr+chr(10)+"Shear: "+str(s)+" Pa - "+sr+chr(10)+"HR: "+str(h)+" bpm"
 
 def tgt_manual(t,p,h,pl,tm):
     risk=sum([float(t)>15,float(p)>2.0,float(h)>50,float(pl)<150])
     return "TAT:"+str(t)+" PF1.2:"+str(p)+chr(10)+"Hemo:"+str(h)+" Plt:"+str(pl)+chr(10)+"RESULT: "+("HIGH RISK" if risk>=3 else "MODERATE" if risk>=2 else "LOW RISK")
 
-
-# ── PHASE D: PROTOCOL GENERATOR + REPORT WRITER + HYPOTHESIS ──────
-
-def generate_protocol(experiment_type, specific_params=""):
-    if not GROQ_KEY: return "Error: Add GROQ_API_KEY to Space Settings."
-    try:
-        client = Groq(api_key=GROQ_KEY)
-        paper_context, _ = search_papers(experiment_type, n=4)
-        lab_context = {
-            "MCL": "Sylgard 184 PDMS 10:1 ratio 48hr cure. Tygon tubing. 70bpm 5L/min 80-120mmHg. Blood analog glycerin water.",
-            "PIV": "Green laser 532nm time-resolved. Seeding particles. Velocity 0.5-2.0 m/s normal. Shear below 5 Pa normal.",
-            "TGT": "Arduino Uno stepper motor 48V. 150mL fresh blood. Sample 0 20 40 60 min. Heska HT5 analyzer.",
-            "uPAD": "Whatman filter paper. Wax printer 120C baking. Picric acid alkaline solution. Jaffe reaction orange-red color.",
-            "FSI": "COMSOL Multiphysics ALE mesh. Blood 1060 kg/m3 0.0035 Pa.s. SJM bileaflet geometry.",
-        }
-        extra = next((v for k,v in lab_context.items() if k.lower() in experiment_type.lower()), "")
-        system = """You are CardioLab AI protocol generator for SJSU Biomedical Engineering.
-Generate a COMPLETE detailed lab protocol. Include these sections:
-1. OBJECTIVE
-2. MATERIALS AND EQUIPMENT (with exact quantities)
-3. SAFETY CONSIDERATIONS
-4. STEP-BY-STEP PROCEDURE (numbered, detailed)
-5. DATA COLLECTION
-6. ANALYSIS METHOD
-7. EXPECTED RESULTS (with normal ranges)
-8. TROUBLESHOOTING
-Use exact SJSU CardioLab values and equipment."""
-        user_msg = f"Generate complete protocol for: {experiment_type}"
-        if specific_params: user_msg += f"
-Parameters: {specific_params}"
-        if extra: user_msg += f"
-CardioLab context: {extra}"
-        if paper_context: user_msg += f"
-From SJSU papers: {paper_context[:600]}"
-        resp = client.chat.completions.create(
-            model="llama-3.3-70b-versatile",
-            messages=[{"role":"system","content":system},{"role":"user","content":user_msg}],
-            max_tokens=1200)
-        return resp.choices[0].message.content
-    except Exception as e: return "Error: "+str(e)
-
-def generate_report(data_description, experiment_type, results=""):
-    if not GROQ_KEY: return "Error: Add GROQ_API_KEY to Space Settings."
-    try:
-        client = Groq(api_key=GROQ_KEY)
-        paper_context, _ = search_papers(experiment_type, n=3)
-        system = """You are CardioLab AI report writer for SJSU Biomedical Engineering.
-Generate a professional research report with these sections:
-1. ABSTRACT (150 words)
-2. INTRODUCTION (background and objectives)
-3. MATERIALS AND METHODS
-4. RESULTS AND DISCUSSION
-5. CONCLUSION
-6. RECOMMENDATIONS
-7. REFERENCES (cite SJSU CardioLab papers)
-Use specific values. Write in professional academic style."""
-        user_msg = f"Write research report for: {experiment_type}"+chr(10)+f"Description: {data_description}"
-        if results: user_msg += chr(10)+f"Results: {results}"
-        if paper_context: user_msg += chr(10)+f"SJSU papers: {paper_context[:600]}"
-        resp = client.chat.completions.create(
-            model="llama-3.3-70b-versatile",
-            messages=[{"role":"system","content":system},{"role":"user","content":user_msg}],
-            max_tokens=1500)
-        return resp.choices[0].message.content
-    except Exception as e: return "Error: "+str(e)
-
-def generate_hypothesis(research_area, current_findings=""):
-    if not GROQ_KEY: return "Error: Add GROQ_API_KEY to Space Settings."
-    try:
-        client = Groq(api_key=GROQ_KEY)
-        paper_context, _ = search_papers(research_area, n=3)
-        system = """You are CardioLab AI research assistant for SJSU Biomedical Engineering.
-Generate 3 specific testable research hypotheses. For each:
-- H0 (null hypothesis)
-- H1 (alternative hypothesis)
-- Scientific rationale
-- Suggested experiment
-- Expected outcome and measurable metrics
-Base on SJSU CardioLab research."""
-        user_msg = f"Generate hypotheses for: {research_area}"
-        if current_findings: user_msg += chr(10)+f"Current findings: {current_findings}"
-        if paper_context: user_msg += chr(10)+f"SJSU papers: {paper_context[:500]}"
-        resp = client.chat.completions.create(
-            model="llama-3.3-70b-versatile",
-            messages=[{"role":"system","content":system},{"role":"user","content":user_msg}],
-            max_tokens=1000)
-        return resp.choices[0].message.content
-    except Exception as e: return "Error: "+str(e)
-
-
+# ── UI ─────────────────────────────────────────────────────────────
 with gr.Blocks(title="CardioLab AI - SJSU", css=CSS) as demo:
     gr.HTML(HEADER)
 
     papers_count = len(set(m["paper"] for m in METADATA)) if PAPERS_LOADED else 0
-    model_status = "CardioLab Fine-tuned Model LOADED" if CARDIOLAB_MODEL_LOADED else "Fine-tuned model loading..."
-    rag_status = f"RAG: {len(CHUNKS)} chunks from {papers_count} SJSU papers" if PAPERS_LOADED else "RAG: loading..."
-    gr.HTML(f'''<div style="background:#1a7340;color:white;text-align:center;padding:7px;font-size:0.82em;font-weight:700;">
-    {rag_status} | {model_status} | Select "CardioLab Fine-tuned (SJSU)" in Model dropdown to use your custom model!</div>''')
+    model_status = "Fine-tuned Model LOADED" if CARDIOLAB_MODEL_LOADED else "Fine-tuned model loading..."
+    rag_status = "RAG: " + str(len(CHUNKS)) + " chunks from " + str(papers_count) + " SJSU papers" if PAPERS_LOADED else "RAG: loading..."
+    gr.HTML("<div style='background:#1a7340;color:white;text-align:center;padding:7px;font-size:0.82em;font-weight:700;'>" + rag_status + " | " + model_status + " | Select CardioLab Fine-tuned in Model dropdown!</div>")
 
     with gr.Tabs():
+
         with gr.Tab("Chat"):
             with gr.Row():
                 with gr.Column(scale=1, min_width=200):
-                    gr.HTML('''<div style="background:#202123;padding:10px;border-radius:8px;margin-bottom:6px;">
-                    <div style="color:#e8a020;font-weight:700;font-size:0.85em;">SJSU CARDIOLAB</div>
-                    <div style="color:#9ca3af;font-size:0.7em;">Conversations</div></div>''')
+                    gr.HTML("<div style='background:#202123;padding:10px;border-radius:8px;margin-bottom:6px;'><div style='color:#e8a020;font-weight:700;font-size:0.85em;'>SJSU CARDIOLAB</div><div style='color:#9ca3af;font-size:0.7em;'>Conversations</div></div>")
                     new_chat_btn = gr.Button("New Chat", variant="secondary")
                     session_dropdown = gr.Dropdown(choices=get_session_list(), label="Saved Sessions", interactive=True)
                     load_btn = gr.Button("Load Session", variant="primary")
@@ -847,14 +683,14 @@ with gr.Blocks(title="CardioLab AI - SJSU", css=CSS) as demo:
             voice_clear.click(lambda: [], outputs=voice_chatbot)
 
         with gr.Tab("Papers"):
-            gr.Markdown("### Search PubMed + Scholar + EuropePMC + ClinicalTrials.gov + SJSU + BioGPT (15M papers)")
+            gr.Markdown("### Search PubMed + Semantic Scholar + SJSU ScholarWorks")
             with gr.Row():
                 search_input = gr.Textbox(placeholder="e.g. bileaflet mechanical heart valve thrombogenicity hemodynamics", label="Research Topic", scale=3)
                 search_model_dd = gr.Dropdown(choices=list(CHAT_MODELS.keys()), value="Llama 3.3 70B (Best)", label="AI Model", scale=1)
                 search_btn = gr.Button("Search", variant="primary", scale=1)
             search_output = gr.Textbox(label="Results", lines=22)
-            search_btn.click(full_research_search, inputs=[search_input, search_model_dd], outputs=search_output)
-            search_input.submit(full_research_search, inputs=[search_input, search_model_dd], outputs=search_output)
+            search_btn.click(quick_search, inputs=[search_input, search_model_dd], outputs=search_output)
+            search_input.submit(quick_search, inputs=[search_input, search_model_dd], outputs=search_output)
 
         with gr.Tab("PIV CSV"):
             with gr.Row():
@@ -930,8 +766,7 @@ with gr.Blocks(title="CardioLab AI - SJSU", css=CSS) as demo:
                                  "uPAD Fabrication","uPAD Creatinine Test",
                                  "FSI COMSOL Simulation","Valve Testing"],
                         value="TGT Blood Testing", label="Experiment Type")
-                    proto_params = gr.Textbox(
-                        placeholder="e.g. 27mm SJM valve 70bpm porcine blood",
+                    proto_params = gr.Textbox(placeholder="e.g. 27mm SJM valve 70bpm porcine blood",
                         label="Specific Parameters", lines=2)
                     proto_btn = gr.Button("Generate Protocol", variant="primary")
                 with gr.Column(scale=2):
@@ -948,7 +783,7 @@ with gr.Blocks(title="CardioLab AI - SJSU", css=CSS) as demo:
                                  "Heart Valve Comparison"],
                         value="TGT Thrombogenicity Study", label="Study Type")
                     report_desc = gr.Textbox(
-                        placeholder="Describe your experiment e.g. TGT with 27mm SJM bileaflet at 70bpm 150mL porcine blood",
+                        placeholder="e.g. TGT with 27mm SJM bileaflet at 70bpm 150mL porcine blood",
                         label="Experiment Description", lines=3)
                     report_results = gr.Textbox(
                         placeholder="e.g. TAT=12.3 ng/mL PF1.2=2.8 Hemo=45 Plt=142",
@@ -978,6 +813,6 @@ with gr.Blocks(title="CardioLab AI - SJSU", css=CSS) as demo:
             hyp_btn.click(generate_hypothesis, inputs=[hyp_area, hyp_findings], outputs=hyp_output)
 
     gr.HTML("""<div style="text-align:center;padding:10px;border-top:1px solid #e5e7eb;background:#f9fafb;">
-    <span style="color:#9ca3af;font-size:0.75em;">CardioLab AI v38 | SJSU Biomedical Engineering | Fine-tuned on 16 SJSU Papers | RAG + Custom Model | Inspired by <a href="https://github.com/snap-stanford/Biomni" style="color:#c1121f;">Biomni Stanford</a> | <a href="https://github.com/pranatechsol/Cardio-Lab-Ai" style="color:#0057a8;">GitHub</a> | Apache 2.0 | $0 Cost</span></div>""")
+    <span style="color:#9ca3af;font-size:0.75em;">CardioLab AI v38 | SJSU Biomedical Engineering | RAG + Fine-tuned + Phase D | Inspired by <a href="https://github.com/snap-stanford/Biomni" style="color:#c1121f;">Biomni Stanford</a> | Apache 2.0 | $0 Cost</span></div>""")
 
 demo.launch()