new UI and hold to speak feature

app.py

@@ -1,57 +1,109 @@
 import streamlit as st
 import plotly.graph_objects as go
-from huggingface_hub import hf_hub_download
-from llama_cpp import Llama
 from faster_whisper import WhisperModel
-# IMPORT THE BACKUP RECORDER
-from audio_recorder_streamlit import audio_recorder
+import requests
 import json
-import json_repair
 import time
 import io
+from audio_recorder_streamlit import audio_recorder
+from PIL import Image
 
 # --- 1. CONFIGURATION ---
 st.set_page_config(page_title="SomAI", layout="wide", page_icon="🩺")
 
-# --- 2. LOAD BRAIN (TURBO MODE) ---
-@st.cache_resource
-def load_models():
-    # A. TEXT BRAIN
-    print(">>> LOADING CORE...")
-    repo_id = "microsoft/Phi-3-mini-4k-instruct-gguf"
-    filename = "Phi-3-mini-4k-instruct-q4.gguf"
-    model_path = hf_hub_download(repo_id=repo_id, filename=filename)
-    
-    llm = Llama(
-        model_path=model_path,
-        n_ctx=1024,
-        n_threads=2,
-        n_batch=512,
-        verbose=False
-    )
-    llm("Hi", max_tokens=1)
+# The URL of your deployed FastAPI backend (Space 2: arshenoy/somAI-backend)
+# NOTE: Replace with the actual URL when deployed. For local testing, use http://localhost:7860
+# When deployed on HF Spaces, this may need to be the actual public URL or a service endpoint if using different Spaces.
+# Assuming the backend is hosted and accessible.
+BACKEND_API_URL = "https://<your-backend-space-name>.hf.space" 
+# Use this for local testing:
+# BACKEND_API_URL = "http://localhost:7860" 
 
-    # B. VOICE BRAIN
+# --- 2. LOAD VOICE BRAIN (Keep Whisper on frontend for VTT) ---
+@st.cache_resource
+def load_whisper():
     print(">>> LOADING AUDIO SENSORS...")
-    whisper = WhisperModel("large-v3", device="cpu", compute_type="int8")
-    
-    return llm, whisper
+    # Use 'tiny' for faster performance on free tier, or 'medium' for better accuracy
+    # 'large-v3' is too slow for frontend VTT in a live chat.
+    whisper_model = WhisperModel("tiny", device="cpu", compute_type="int8") 
+    return whisper_model
 
 try:
-    llm, whisper = load_models()
+    whisper = load_whisper()
 except Exception as e:
-    st.error(f"SYSTEM FAILURE: {e}")
+    st.error(f"WHISPER FAILURE: {e}")
     st.stop()
 
-# --- 3. WHATSAPP STYLE CSS ---
-st.markdown("""<style>
-    .stApp {background-color: #0e1117;}
-    section[data-testid="stSidebar"] {background-color: #161b22; border-right: 1px solid #30363d;}
-    .metric-card {background-color: #161b22; border: 1px solid #30363d; border-radius: 8px; padding: 15px; margin-bottom: 10px;}
-    .chat-bubble {padding: 12px 16px; border-radius: 8px; margin-bottom: 10px; line-height: 1.5; font-family: sans-serif; font-size: 16px;}
-    .user-bubble {background-color: #238636; color: #ffffff !important; margin-left: 20%; border-radius: 12px 12px 0 12px; text-align: right;}
-    .ai-bubble {background-color: #21262d; border: 1px solid #30363d; color: #ffffff !important; margin-right: 20%; border-radius: 12px 12px 12px 0; text-align: left;}
-</style>""", unsafe_allow_html=True)
+# --- 3. NEW NEON STYLE CSS ---
+st.markdown("""
+<style>
+    @import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;600&family=JetBrains+Mono:wght@400;700&display=swap');
+
+    /* BASE THEME - Deep Space Black */
+    .stApp {
+        background-color: #050505;
+        background-image: radial-gradient(circle at 50% 50%, #1a1a1a 0%, #000000 100%);
+        font-family: 'Inter', sans-serif;
+    }
+
+    /* GLASS SIDEBAR */
+    section[data-testid="stSidebar"] {
+        background: rgba(10, 10, 10, 0.7);
+        backdrop-filter: blur(12px);
+        border-right: 1px solid rgba(255, 255, 255, 0.08);
+    }
+
+    /* NEON METRICS */
+    div[data-testid="metric-container"] {
+        background: rgba(255, 255, 255, 0.03);
+        border: 1px solid rgba(255, 255, 255, 0.1);
+        border-radius: 12px;
+        padding: 15px;
+        transition: 0.3s;
+    }
+    div[data-testid="metric-container"]:hover {
+        border-color: #00ff80;
+        box-shadow: 0 0 15px rgba(0, 255, 128, 0.1);
+    }
+
+    /* TEXT GLOW */
+    h1, h2, h3 {
+        font-family: 'JetBrains Mono', monospace;
+        letter-spacing: -0.5px;
+        color: #fff;
+        text-shadow: 0 0 10px rgba(255, 255, 255, 0.2);
+    }
+    
+    /* CHAT BUBBLES - Updated for new dark background */
+    .chat-bubble {
+        padding: 12px 16px;
+        border-radius: 8px;
+        margin-bottom: 10px;
+        line-height: 1.5;
+        font-family: 'Inter', sans-serif; 
+        font-size: 16px;
+        color: #e0e0e0;
+    }
+    .user-bubble {
+        background-color: #004d26; /* Darker green for user */
+        margin-left: 20%; 
+        border-radius: 12px 12px 0 12px; 
+        text-align: right;
+    }
+    .ai-bubble {
+        background-color: #1a1a1a; /* Dark gray for AI */
+        border: 1px solid rgba(255, 255, 255, 0.1);
+        margin-right: 20%; 
+        border-radius: 12px 12px 12px 0; 
+        text-align: left;
+    }
+
+    /* REMOVE JUNK */
+    #MainMenu {visibility: hidden;}
+    footer {visibility: hidden;}
+</style>
+""", unsafe_allow_html=True)
+
 
 # --- 4. SESSION STATE ---
 if 'history' not in st.session_state: st.session_state.history = []
@@ -61,129 +113,198 @@ if 'mode' not in st.session_state: st.session_state.mode = "GENERAL"
 
 # --- 5. SIDEBAR ---
 with st.sidebar:
-    st.markdown("### Patient Intake & Vitals")
+    st.markdown("### 🧬 Patient Intake & Vitals")
     with st.expander("Patient Profile", expanded=True):
         name = st.text_input("Name", "Patient X")
         age = st.slider("Age", 18, 90, 20)
-            
+        condition = st.text_input("Primary Condition", "Diabetes/Hypertension")
+                
     with st.expander("Clinical Vitals", expanded=True):
         bp = st.number_input("Systolic BP", 90, 220, 110)
         glucose = st.number_input("Glucose", 70, 400, 110)
-        sleep = st.slider("Sleep", 0, 10, 4)
-        meds = st.slider("Missed Doses", 0, 7, 3)
+        sleep = st.slider("Sleep Quality (0-10)", 0, 10, 4)
+        meds = st.slider("Missed Doses (Last 7 Days)", 0, 7, 3)
 
     if st.button("RUN CLINICAL ANALYSIS", type="primary", use_container_width=True):
-        with st.spinner("Processing..."):
-            score = 10
-            if bp > 140: score += 20
-            if glucose > 180: score += 20
-            if sleep < 5: score += 15
-            if meds > 2: score += 20
-            score = min(score, 100)
-            
-            sys = "Medical Analyst. Output strictly valid JSON."
-            user_data = f"Age: {age}, BP: {bp}, Glucose: {glucose}, Risk: {score}/100."
-            task = """Task: 1. Risk Summary. 2. 3 Actions. Format: {"summary": "string", "actions": ["a", "b", "c"]}"""
-            prompt = f"<|system|>{sys}<|end|><|user|>{user_data}\n{task}<|end|><|assistant|>"
-            
-            output = llm(prompt, max_tokens=200, temperature=0.1, stop=["<|end|>", "###"])
+        with st.spinner("Analyzing Clinical Markers..."):
             try:
-                data = json_repair.repair_json(output['choices'][0]['text'], return_objects=True)
-                st.session_state.risk_score = score
-                st.session_state.risk_summary = data.get('summary', "Review vitals.")
-            except:
-                st.session_state.risk_score = score
-                st.session_state.risk_summary = "Analysis Complete."
+                # API CALL TO /analyze ENDPOINT
+                payload = {
+                    "age": age,
+                    "condition": condition,
+                    "sleep_quality": sleep,
+                    "missed_doses": meds,
+                    "systolic_bp": bp,
+                    "glucose": glucose
+                }
+                response = requests.post(f"{BACKEND_API_URL}/analyze", json=payload, timeout=30)
+                response.raise_for_status() # Raises an HTTPError for bad responses (4xx or 5xx)
+                
+                data = response.json()
+                st.session_state.risk_score = data['numeric_score']
+                st.session_state.risk_summary = data['risk_summary']
+            
+            except requests.exceptions.RequestException as req_err:
+                st.error(f"API Error: Cannot connect to backend (Code: {req_err.response.status_code if hasattr(req_err, 'response') and req_err.response else 'N/A'}). Ensure Space 2 is running.")
+                st.session_state.risk_score = 0
+                st.session_state.risk_summary = "Backend service unavailable."
+            except Exception as e:
+                st.error(f"Analysis Failed: {e}")
+                st.session_state.risk_score = 0
+                st.session_state.risk_summary = "Processing error."
+
 
     val = st.session_state.risk_score
-    color = "#238636" if val < 40 else "#e3b341" if val < 80 else "#da3633"
-    fig = go.Figure(go.Indicator(mode="gauge+number", value=val, number={'font': {'size': 40, 'color': 'white'}}, gauge={'axis': {'range': [0, 100]}, 'bar': {'color': color}, 'bgcolor': "#0d1117", 'bordercolor': "#30363d"}))
-    fig.update_layout(height=250, margin=dict(l=10,r=10,t=30,b=10), paper_bgcolor="rgba(0,0,0,0)")
+    color = "#00ff80" if val < 40 else "#ffc300" if val < 80 else "#ff3300" # Neon color scheme
+
+    # Gauge Chart
+    fig = go.Figure(go.Indicator(
+        mode="gauge+number", 
+        value=val, 
+        number={'font': {'size': 40, 'color': 'white'}}, 
+        gauge={
+            'axis': {'range': [0, 100], 'tickcolor': '#333333'}, 
+            'bar': {'color': color}, 
+            'bgcolor': "rgba(26, 26, 26, 0.7)", 
+            'bordercolor': "#333333",
+            'steps': [
+                {'range': [0, 40], 'color': 'rgba(0, 255, 128, 0.1)'},
+                {'range': [40, 80], 'color': 'rgba(255, 195, 0, 0.1)'},
+                {'range': [80, 100], 'color': 'rgba(255, 51, 0, 0.1)'},
+            ]
+        }
+    ))
+    fig.update_layout(
+        height=250, 
+        margin=dict(l=10,r=10,t=30,b=10), 
+        paper_bgcolor="rgba(0,0,0,0)", 
+        font={'color': 'white', 'family': 'JetBrains Mono'}
+    )
     st.plotly_chart(fig, use_container_width=True)
-    st.markdown(f"""<div class="metric-card" style="border-left: 4px solid {color}"><h5 style="margin:0; color: {color}">Clinical Assessment</h5><p style="margin-top:5px; font-size: 0.9rem; color: #ddd;">{st.session_state.risk_summary}</p></div>""", unsafe_allow_html=True)
+    
+    # Custom Metric Card with Border (replaces standard metric-card)
+    st.markdown(f"""
+    <div style="background: rgba(255, 255, 255, 0.05); border: 1px solid rgba(255, 255, 255, 0.1); border-radius: 12px; padding: 15px; margin-top: 15px; border-left: 4px solid {color};">
+        <h5 style="margin:0; color: {color}; font-family: 'JetBrains Mono', monospace;">CLINICAL ASSESSMENT</h5>
+        <p style="margin-top:5px; font-size: 0.9rem; color: #ddd;">{st.session_state.risk_summary}</p>
+    </div>
+    """, unsafe_allow_html=True)
 
 # --- 6. MAIN CHAT ---
 col_main = st.columns(1)[0] 
 with col_main:
-    st.markdown("## SomAI Assistant")
+    st.markdown("## 🧠 SomAI Assistant")
     c1, c2 = st.columns(2)
-    if c1.button("Medical Guide", use_container_width=True, type="primary" if st.session_state.mode == "GENERAL" else "secondary"): st.session_state.mode = "GENERAL"
-    if c2.button("Therapist", use_container_width=True, type="primary" if st.session_state.mode == "THERAPY" else "secondary"): st.session_state.mode = "THERAPY"
-
+    if c1.button("🩺 Medical Guide", use_container_width=True, type="primary" if st.session_state.mode == "GENERAL" else "secondary"): st.session_state.mode = "GENERAL"
+    if c2.button("🫂 Therapist", use_container_width=True, type="primary" if st.session_state.mode == "THERAPY" else "secondary"): st.session_state.mode = "THERAPY"
+    
+    # Display History
     chat_container = st.container(height=400)
     for msg in st.session_state.history:
         div_class = "user-bubble" if msg['role'] == "user" else "ai-bubble"
         chat_container.markdown(f"<div class='chat-bubble {div_class}'>{msg['content']}</div>", unsafe_allow_html=True)
-
-    # --- ROBUST AUDIO INPUT (The Fix) ---
-    st.write("🎙️ **Voice Input:**")
+    
+    # --- ROBUST AUDIO INPUT (Hold and Speak) ---
+    st.markdown("---")
+    st.markdown("🎙️ **Hold & Speak:**")
     audio_bytes = audio_recorder(
         text="",
-        recording_color="#ff0000",
-        neutral_color="#ffffff",
+        recording_color="#ff3300", # Neon Red for recording
+        neutral_color="#00ff80",    # Neon Green for neutral
         icon_name="microphone",
-        icon_size="2x",
+        icon_size="3x",
+        initial_time=0 # Force 'hold-to-speak' mode
     )
     
     user_query = None
     
     # 1. VOICE PROCESSING
     if audio_bytes:
-        # Save bytes to a virtual file so Whisper can read it
-        with st.spinner("Processing Voice..."):
+        with st.spinner("🔊 Transcribing Voice..."):
             audio_file = io.BytesIO(audio_bytes)
-            # Create a simplified audio object or save temp
-            # Whisper usually needs a file path or file-like object
-            segments, info = whisper.transcribe(audio_file, beam_size=1)
+            # Transcribe the audio
+            # Using a single file-like object directly with WhisperModel
+            segments, info = whisper.transcribe(audio_file, beam_size=5) 
             text_list = [segment.text for segment in segments]
-            user_query = " ".join(text_list)
+            user_query = " ".join(text_list).strip()
+            # If transcription is empty, handle gracefully
+            if not user_query:
+                st.warning("Could not detect speech. Please speak clearly.")
+                st.stop()
+            st.session_state.history.append({"role": "user", "content": user_query})
+            chat_container.markdown(f"<div class='chat-bubble user-bubble'>{user_query}</div>", unsafe_allow_html=True)
     
     # 2. TEXT PROCESSING
     text_input = st.chat_input("...or type a message")
     if text_input:
         user_query = text_input
-
-    # 3. QUERY LOGIC
-    if user_query:
         st.session_state.history.append({"role": "user", "content": user_query})
         chat_container.markdown(f"<div class='chat-bubble user-bubble'>{user_query}</div>", unsafe_allow_html=True)
+    
+    # 3. QUERY LOGIC (API Call)
+    if user_query:
+        
+        # --- LLM Response Generation (Streaming Emulation) ---
+        placeholder = chat_container.empty()
+        full_resp = ""
+        
+        with placeholder.container():
+            with st.spinner("Thinking..."):
+                try:
+                    # API Call to /generate ENDPOINT
+                    payload = {
+                        "query": user_query,
+                        "age": age,
+                        "condition": condition,
+                        "mode": st.session_state.mode
+                    }
+                    
+                    # Using Stream=False for FastAPI/requests.post since the current FastAPI code doesn't support streaming.
+                    # We will stream the *display* locally to emulate the effect.
+                    response = requests.post(f"{BACKEND_API_URL}/generate", json=payload, timeout=60)
+                    response.raise_for_status() 
+                    
+                    data = response.json()
+                    raw_text = data['generated_text']
+                    
+                    # Stream display emulation: Chunk the text and write it
+                    # This simulates streaming visually even if the API is non-streaming.
+                    chunk_size = 5 # words per chunk
+                    words = raw_text.split()
+                    
+                    for i in range(0, len(words), chunk_size):
+                        chunk = " ".join(words[i:i + chunk_size])
+                        full_resp += chunk + " "
+                        
+                        # Use an empty container to display the response with a cursor effect
+                        # and then replace it with the next chunk
+                        placeholder.markdown(f"<div class='chat-bubble ai-bubble'>{full_resp}▌</div>", unsafe_allow_html=True)
+                        time.sleep(0.05) # Adjust for speed
+                    
+                    # Final display (no cursor)
+                    placeholder.markdown(f"<div class='chat-bubble ai-bubble'>{raw_text}</div>", unsafe_allow_html=True)
+                    
+                    st.session_state.history.append({"role": "assistant", "content": raw_text})
+                    
+                    # Display suggestions below the chat container
+                    suggestions = data.get('suggestions', [])
+                    if suggestions:
+                        st.markdown("---")
+                        st.markdown("💡 **Next Steps:**")
+                        suggestion_cols = st.columns(len(suggestions))
+                        for i, sug in enumerate(suggestions):
+                            suggestion_cols[i].button(sug, key=f"sug_{i}_{len(st.session_state.history)}", use_container_width=True)
 
-        prompt_lower = user_query.lower().strip()
-
-        # CHEATS
-        if "demo chest pain" in prompt_lower:
-            response = "**ANALYSIS:** Potential **Angina Pectoris**.\n\n**URGENT:**\n1. Radiation to arm?\n2. Shortness of breath?\n\n**ADVICE:** Proceed to ER for ECG."
-            with chat_container:
-                with st.spinner("Analyzing..."):
-                    time.sleep(1)
-                chat_container.markdown(f"<div class='chat-bubble ai-bubble'>{response}</div>", unsafe_allow_html=True)
-            st.session_state.history.append({"role": "assistant", "content": response})
-            st.stop()
-            
-        if prompt_lower in ["hi", "hello", "hey"]:
-            resp = "Hello! I am SomAI. Describe your symptoms."
-            st.session_state.history.append({"role": "assistant", "content": resp})
-            chat_container.markdown(f"<div class='chat-bubble ai-bubble'>{resp}</div>", unsafe_allow_html=True)
-            st.stop()
-
-        if any(x in prompt_lower for x in ["die", "suicide", "kill"]):
-            resp = "🚨 Safety Protocol Engaged. Contact 988."
-            st.session_state.history.append({"role": "assistant", "content": resp})
-            chat_container.markdown(f"<div class='chat-bubble ai-bubble' style='border-color:red'>{resp}</div>", unsafe_allow_html=True)
-            st.stop()
-
-        # REAL AI
-        sys = "Medical Advisor. Ask 1 follow-up. Be concise." if st.session_state.mode == "GENERAL" else "Therapist. Be supportive."
-        full_prompt = f"<|system|>{sys}<|end|><|user|>{user_query}<|end|><|assistant|>"
-
-        with chat_container:
-            placeholder = st.empty()
-            full_resp = ""
-            stream = llm(full_prompt, max_tokens=300, stream=True, stop=["<|end|>", "###"])
-            for chunk in stream:
-                txt = chunk['choices'][0]['text']
-                full_resp += txt
-                placeholder.markdown(f"<div class='chat-bubble ai-bubble'>{full_resp}▌</div>", unsafe_allow_html=True)
-            placeholder.markdown(f"<div class='chat-bubble ai-bubble'>{full_resp}</div>", unsafe_allow_html=True)
+                except requests.exceptions.RequestException as req_err:
+                    error_msg = f"API Error: {req_err}. Check backend service health."
+                    st.error(error_msg)
+                    st.session_state.history.append({"role": "assistant", "content": error_msg})
+                    placeholder.markdown(f"<div class='chat-bubble ai-bubble'>{error_msg}</div>", unsafe_allow_html=True)
+                except Exception as e:
+                    error_msg = f"LLM Generation Failed: {e}"
+                    st.error(error_msg)
+                    st.session_state.history.append({"role": "assistant", "content": error_msg})
+                    placeholder.markdown(f"<div class='chat-bubble ai-bubble'>{error_msg}</div>", unsafe_allow_html=True)
         
-        st.session_state.history.append({"role": "assistant", "content": full_resp})
+        # Manually rerun to clear the input box and update history
+        st.rerun()