Refactor: Consolidate AI Investigation into single unified agent tab with RAG capabilities

- Replace 4 separate investigation tabs (Vehicle, Area, Midnight, Most Suspicious) with 1 unified tab
- Add ask_investigation_question() function for natural language queries with RAG
- Agent now dynamically understands questions and fetches relevant database data
- Users can ask any question (e.g. 'vehicles in Adyar at midnight', 'suspicious activity in industrial zones')
- Mistral model acts as RAG agent to analyze data and provide intelligent answers
- Add example quick-start buttons for common investigation types

ai_investigation.py

@@ -536,6 +536,283 @@ def find_most_suspicious_today():
 # FORMAT OUTPUT
 # =====================================================
 
+# =====================================================
+# UNIFIED AI AGENT - NATURAL LANGUAGE QUESTIONS
+# =====================================================
+
+def ask_investigation_question(question):
+    """
+    Unified AI Investigation Agent
+    
+    Takes any natural language question and:
+    1. Uses Mistral to understand what data is needed
+    2. Queries the database based on the question intent
+    3. Uses Mistral to generate intelligent answer with RAG
+    """
+    
+    if not question or len(question.strip()) < 3:
+        return {
+            "status": "error",
+            "message": "Please enter a valid question"
+        }
+    
+    client = get_mistral_client()
+    
+    # Step 1: Understand query intent
+    intent_prompt = f"""Analyze this surveillance question and determine what type of data is needed.
+    
+Question: {question}
+
+Respond with ONLY one of these intents in the format:
+INTENT: [type]
+QUERY: [specific value or description]
+
+Types and examples:
+1. VEHICLE - plate number like "TN12UP3854" or "find vehicle in [location]"
+2. LOCATION - area name like "Adyar", "Nungambakkam", "Guindy"
+3. MIDNIGHT - late night activity in specific area or general
+4. ACTIVITY - activity patterns in area during specific times
+5. SUSPICIOUS - find suspicious vehicles matching criteria
+6. TIME_RANGE - activity in specific time period or area
+
+Extract the specific target from the question (plate, location, time, etc)."""
+
+    try:
+        if client:
+            intent_response = client.chat(
+                model="mistral-small",
+                messages=[{"role": "user", "content": intent_prompt}],
+                temperature=0.3,
+                max_tokens=100
+            )
+            intent_text = intent_response.choices[0].message.content
+        else:
+            intent_text = _parse_intent_fallback(question)
+        
+        print(f"Intent parsed: {intent_text}")
+        
+        # Step 2: Execute query based on intent
+        data_context = _execute_intelligent_query(intent_text, question)
+        
+        # Step 3: Generate answer with RAG
+        answer = _generate_rag_answer(question, data_context, client)
+        
+        return {
+            "status": "success",
+            "question": question,
+            "answer": answer,
+            "data_summary": data_context.get("summary", ""),
+            "analysis": data_context.get("analysis", {}),
+            "findings": data_context.get("findings", [])
+        }
+        
+    except Exception as e:
+        print(f"Error in investigation: {e}")
+        return {
+            "status": "error",
+            "message": f"Investigation failed: {str(e)}"
+        }
+
+
+def _execute_intelligent_query(intent_text, question):
+    """Execute database query based on parsed intent"""
+    
+    intent_text = intent_text.upper()
+    data = {
+        "summary": "",
+        "analysis": {},
+        "findings": [],
+        "raw_data": None
+    }
+    
+    try:
+        # Parse intent and query
+        if "INTENT: VEHICLE" in intent_text:
+            # Extract plate from question or intent
+            plate = _extract_plate_from_text(intent_text + " " + question)
+            if plate:
+                df = get_vehicle_detections(plate, days=30)
+                if not df.empty:
+                    analysis = analyze_vehicle_behavior(df)
+                    data["analysis"] = analysis
+                    data["summary"] = f"Found {len(df)} detections for vehicle {plate}"
+                    data["findings"] = _extract_key_findings({"analysis": analysis})
+                    data["raw_data"] = df.to_dict(orient="records")[:10]
+        
+        elif "INTENT: LOCATION" in intent_text:
+            # Extract location
+            location = _extract_location_from_text(intent_text + " " + question)
+            if location:
+                df = get_area_activity(location, days=7)
+                if not df.empty:
+                    analysis = analyze_area_patterns(df)
+                    data["analysis"] = analysis
+                    data["summary"] = f"Found {len(df)} detections in {location}"
+                    data["findings"] = _extract_key_findings({"analysis": analysis})
+                    data["raw_data"] = df.to_dict(orient="records")[:10]
+        
+        elif "INTENT: MIDNIGHT" in intent_text:
+            # Extract location if mentioned
+            location = _extract_location_from_text(question)
+            df = get_midnight_activity(hours_start=22, hours_end=4, days=7)
+            
+            if location:
+                df = df[df["location"].str.contains(location, case=False, na=False)]
+            
+            if not df.empty:
+                analysis = analyze_midnight_patterns(df)
+                data["analysis"] = analysis
+                data["summary"] = f"Found {len(df)} midnight detections"
+                data["findings"] = _extract_key_findings({"analysis": analysis})
+                data["raw_data"] = df.to_dict(orient="records")[:10]
+        
+        elif "INTENT: ACTIVITY" in intent_text:
+            # Get activity in specific area
+            location = _extract_location_from_text(question)
+            if location:
+                df = get_area_activity(location, days=7)
+                if not df.empty:
+                    analysis = analyze_area_patterns(df)
+                    data["analysis"] = analysis
+                    data["summary"] = f"Analyzed activity in {location}"
+                    data["findings"] = _extract_key_findings({"analysis": analysis})
+                    data["raw_data"] = df.to_dict(orient="records")[:10]
+        
+        if not data["summary"]:
+            # Fallback: get general statistics
+            query = "SELECT COUNT(*) as total_detections, COUNT(DISTINCT plate) as unique_vehicles, COUNT(DISTINCT location) as unique_locations FROM vehicle_logs WHERE date >= CURRENT_DATE - INTERVAL '7 days'"
+            with engine.connect() as conn:
+                result = conn.execute(text(query))
+                row = result.fetchone()
+                if row:
+                    data["summary"] = f"Database has {row[0]} total detections, {row[1]} unique vehicles, {row[2]} unique locations"
+    
+    except Exception as e:
+        print(f"Error executing query: {e}")
+        data["summary"] = "Unable to fetch data from database"
+    
+    return data
+
+
+def _generate_rag_answer(question, data_context, client):
+    """Generate RAG-based answer using Mistral"""
+    
+    context_str = json.dumps(data_context, indent=2, default=str)
+    
+    rag_prompt = f"""You are a professional surveillance and traffic analysis AI agent.
+    
+User Question: {question}
+
+Available Data:
+{context_str}
+
+Generate a detailed, professional answer to the user's question using the available data.
+
+Guidelines:
+1. Answer directly and concisely
+2. Use the data to provide specific insights
+3. Highlight patterns, anomalies, and recommendations
+4. Be factual and data-driven
+5. If data is limited, explain what additional data would help
+
+Format your response in a clear, organized manner with sections if needed."""
+
+    try:
+        if client:
+            response = client.chat(
+                model="mistral-small",
+                messages=[{"role": "user", "content": rag_prompt}],
+                temperature=0.3,
+                max_tokens=800
+            )
+            return response.choices[0].message.content
+        else:
+            return _generate_fallback_rag_answer(question, data_context)
+    
+    except Exception as e:
+        print(f"Error generating answer: {e}")
+        return _generate_fallback_rag_answer(question, data_context)
+
+
+def _generate_fallback_rag_answer(question, data_context):
+    """Fallback answer generation without Mistral"""
+    
+    answer = f"**Analysis for:** {question}\n\n"
+    answer += f"**Summary:** {data_context.get('summary', 'No data available')}\n\n"
+    
+    if data_context.get("analysis"):
+        answer += "**Key Metrics:**\n"
+        for key, value in list(data_context["analysis"].items())[:5]:
+            answer += f"• {key}: {value}\n"
+    
+    if data_context.get("findings"):
+        answer += "\n**Findings:**\n"
+        for finding in data_context["findings"]:
+            answer += f"• {finding}\n"
+    
+    return answer
+
+
+def _extract_plate_from_text(text):
+    """Extract license plate from text"""
+    import re
+    # Look for plate patterns like "TN12UP3854" or "TN 12 UP 3854"
+    plate_patterns = [
+        r'([A-Z]{2}\d{2}[A-Z]{2}\d{4})',
+        r'([A-Z]{2} \d{2} [A-Z]{2} \d{4})',
+    ]
+    
+    for pattern in plate_patterns:
+        match = re.search(pattern, text.upper())
+        if match:
+            return match.group(1).replace(" ", "")
+    
+    return None
+
+
+def _extract_location_from_text(text):
+    """Extract location from text"""
+    # Common locations in Chennai
+    locations = [
+        "adyar", "anna nagar", "nungambakkam", "guindy", "chetpet",
+        "tnagar", "t nagar", "mylapore", "velachery", "tambaram",
+        "chromepet", "madhavaram", "tiruvottiyur", "ayanavaram",
+        "anna salai", "mg road", "mount road", "geek city",
+        "industrial estate", "delphi", "siruseri", "sholinganallur"
+    ]
+    
+    text_lower = text.lower()
+    for location in locations:
+        if location in text_lower:
+            return location.title()
+    
+    # Try to find any capitalized word that might be a location
+    import re
+    words = re.findall(r'\b[A-Z][a-z]+(?:\s[A-Z][a-z]+)*\b', text)
+    if words:
+        return words[-1]
+    
+    return None
+
+
+def _parse_intent_fallback(question):
+    """Fallback intent parsing without Mistral"""
+    q_lower = question.lower()
+    
+    if any(word in q_lower for word in ["vehicle", "plate", "car", "auto"]):
+        return "INTENT: VEHICLE"
+    elif any(word in q_lower for word in ["midnight", "night", "late", "early morning", "2am", "3am"]):
+        return "INTENT: MIDNIGHT"
+    elif any(word in q_lower for word in ["area", "location", "place", "zone", "street", "road"]):
+        return "INTENT: LOCATION"
+    elif any(word in q_lower for word in ["activity", "movement", "traffic", "busy"]):
+        return "INTENT: ACTIVITY"
+    elif any(word in q_lower for word in ["suspicious", "alert", "danger", "risk", "threat"]):
+        return "INTENT: SUSPICIOUS"
+    else:
+        return "INTENT: ACTIVITY"
+
+
 def format_investigation_output(result):
     """Format investigation result for display"""
     if result.get("status") == "error":
@@ -544,11 +821,12 @@ def format_investigation_output(result):
         }
     
     return {
-        "query": result.get("plate") or result.get("location") or "Midnight Activity",
+        "query": result.get("plate") or result.get("location") or "Investigation",
         "investigation_narrative": result.get("investigation_narrative", ""),
+        "answer": result.get("answer", ""),
         "analysis": result.get("analysis", {}),
-        "key_findings": _extract_key_findings(result),
-        "recommendations": _extract_recommendations(result.get("investigation_narrative", ""))
+        "key_findings": result.get("findings", []),
+        "recommendations": _extract_recommendations(result.get("investigation_narrative", "") or result.get("answer", ""))
     }
 
 

app.py

@@ -40,6 +40,7 @@ from ai_investigation import (
     investigate_area,
     investigate_midnight_activity,
     find_most_suspicious_today,
+    ask_investigation_question,
     format_investigation_output
 )
 
@@ -99,6 +100,56 @@ def investigate_area_wrapper(location):
     )
 
 
+def investigate_midnight_wrapper():
+    """Wrapper for midnight investigation"""
+    result = investigate_midnight_activity()
+    if result.get("status") == "error":
+        error_msg = result.get("message", "Investigation failed")
+        return (f"❌ Error: {error_msg}", "", {}, "")
+    
+    output = format_investigation_output(result)
+    return (
+        output.get("investigation_narrative", ""),
+        "\n".join(output.get("key_findings", [])),
+        output.get("analysis", {}),
+        "\n".join(output.get("recommendations", []))
+    )
+
+
+def find_suspicious_wrapper():
+    """Wrapper for finding most suspicious today"""
+    result = find_most_suspicious_today()
+    if result.get("status") == "error":
+        error_msg = result.get("message", "Investigation failed")
+        return (f"❌ Error: {error_msg}", "", {}, "")
+    
+    output = format_investigation_output(result)
+    return (
+        output.get("investigation_narrative", ""),
+        "\n".join(output.get("key_findings", [])),
+        output.get("analysis", {}),
+        "\n".join(output.get("recommendations", []))
+    )
+
+
+def ask_question_wrapper(question):
+    """Unified AI Investigation Agent - handles any natural language question"""
+    if not question or len(question.strip()) < 3:
+        return ("Please ask a valid question about vehicles, locations, or activities", "", {}, "")
+    
+    result = ask_investigation_question(question)
+    if result.get("status") == "error":
+        error_msg = result.get("message", "Investigation failed")
+        return (f"❌ Error: {error_msg}", "", {}, "")
+    
+    return (
+        result.get("answer", ""),
+        "\n".join(result.get("findings", [])),
+        result.get("analysis", {}),
+        ""
+    )
+
+
 def investigate_midnight_wrapper():
     """Wrapper for midnight activity investigation"""
     result = investigate_midnight_activity()
@@ -623,181 +674,93 @@ with gr.Blocks(
 
 
     # =====================================================
-    # TAB 3: AI Assistant
+    # TAB 3: AI Assistant - Unified Investigation Agent
     # =====================================================
 
     with gr.Tab("🔍 AI Investigation", id="tab_investigation"):
         
         gr.Markdown("""
-        # 🔍 AI Investigation Assistant
+        # 🔍 AI Investigation Assistant - Smart Agent with RAG
+        
+        Ask any question about vehicles, locations, or activities. The AI agent will:
+        - 🤖 Understand your question
+        - 📊 Fetch relevant data from the database
+        - 🧠 Analyze patterns and provide intelligent insights
+        - 📋 Generate detailed investigation reports
         
-        Professional surveillance analysis system. Analyzes vehicle behavior patterns and generates intelligence narratives.
+        **Examples of questions you can ask:**
+        - "What vehicles were in Adyar at midnight?"
+        - "Show me suspicious activity in Nungambakkam between 10PM and 3AM"
+        - "Which vehicles are most active in industrial zones?"
+        - "What's the activity level in Guindy today?"
+        - "Find vehicles with multiple detections in different areas"
+        - "Which plates appear most frequently at night?"
+        - "Analyze traffic patterns in Anna Nagar"
         """)
-
-        with gr.Tabs():
-            
-            # =====================================================
-            # INVESTIGATION TYPE 1: VEHICLE
-            # =====================================================
-            
-            with gr.Tab("🚗 Vehicle Investigation"):
-                gr.Markdown("### Investigate single vehicle behavior")
-                
-                invest_plate = gr.Textbox(
-                    placeholder="TN12UP3854",
-                    label="License Plate",
-                    scale=2
-                )
-                
-                investigate_btn = gr.Button("🔎 Investigate Vehicle", variant="primary")
-                
-                with gr.Row():
-                    invest_narrative = gr.Textbox(
-                        label="📋 Investigation Narrative",
-                        lines=8,
-                        interactive=False
-                    )
-                    
-                    invest_findings = gr.Textbox(
-                        label="🚨 Key Findings",
-                        lines=8,
-                        interactive=False
-                    )
-                
-                invest_analysis = gr.JSON(
-                    label="📊 Detailed Analysis"
-                )
-                
-                invest_recommendations = gr.Textbox(
-                    label="✅ Recommendations",
-                    lines=6,
-                    interactive=False
-                )
-                
-                investigate_btn.click(
-                    fn=investigate_vehicle_wrapper,
-                    inputs=[invest_plate],
-                    outputs=[invest_narrative, invest_findings, invest_analysis, invest_recommendations]
-                )
-            
-            # =====================================================
-            # INVESTIGATION TYPE 2: AREA
-            # =====================================================
+        
+        with gr.Row():
+            question_input = gr.Textbox(
+                placeholder="e.g., What vehicles were detected in Adyar at midnight? Or: Show me suspicious activity in industrial zones tonight",
+                label="🤔 Ask Any Question About Vehicles, Locations, or Activities",
+                lines=3,
+                scale=5
+            )
             
-            with gr.Tab("📍 Area Intelligence"):
-                gr.Markdown("### Analyze activity in specific area/location")
-                
-                invest_location = gr.Textbox(
-                    placeholder="Nungambakkam_HighRoad or Guindy",
-                    label="Location/Area Name",
-                    scale=2
-                )
-                
-                area_investigate_btn = gr.Button("🔎 Analyze Area", variant="primary")
-                
-                with gr.Row():
-                    area_narrative = gr.Textbox(
-                        label="📋 Area Intelligence Report",
-                        lines=8,
-                        interactive=False
-                    )
-                    
-                    area_findings = gr.Textbox(
-                        label="🚨 Key Findings",
-                        lines=8,
-                        interactive=False
-                    )
-                
-                area_analysis = gr.JSON(
-                    label="📊 Area Analysis"
-                )
-                
-                area_recommendations = gr.Textbox(
-                    label="✅ Security Recommendations",
-                    lines=6,
+            investigate_btn = gr.Button("🔎 Investigate", variant="primary", scale=1, size="lg")
+        
+        # Output sections
+        with gr.Tabs():
+            with gr.Tab("📋 AI Response"):
+                response_output = gr.Textbox(
+                    label="🤖 AI Investigation Analysis",
+                    lines=12,
                     interactive=False
                 )
-                
-                area_investigate_btn.click(
-                    fn=investigate_area_wrapper,
-                    inputs=[invest_location],
-                    outputs=[area_narrative, area_findings, area_analysis, area_recommendations]
-                )
             
-            # =====================================================
-            # INVESTIGATION TYPE 3: MIDNIGHT ACTIVITY
-            # =====================================================
-            
-            with gr.Tab("🌙 Midnight Activity"):
-                gr.Markdown("### Find suspicious midnight/late-night movements")
-                
-                midnight_btn = gr.Button("🔎 Analyze Midnight Activity", variant="primary", scale=2)
-                
-                with gr.Row():
-                    midnight_narrative = gr.Textbox(
-                        label="📋 Night Activity Report",
-                        lines=8,
-                        interactive=False
-                    )
-                    
-                    midnight_findings = gr.Textbox(
-                        label="🚨 Suspicious Patterns",
-                        lines=8,
-                        interactive=False
-                    )
-                
-                midnight_analysis = gr.JSON(
-                    label="📊 Night Analysis"
-                )
-                
-                midnight_recommendations = gr.Textbox(
-                    label="✅ Tactical Recommendations",
-                    lines=6,
+            with gr.Tab("🚨 Key Findings"):
+                findings_output = gr.Textbox(
+                    label="Key Findings & Anomalies",
+                    lines=10,
                     interactive=False
                 )
-                
-                midnight_btn.click(
-                    fn=investigate_midnight_wrapper,
-                    inputs=[],
-                    outputs=[midnight_narrative, midnight_findings, midnight_analysis, midnight_recommendations]
-                )
-            
-            # =====================================================
-            # INVESTIGATION TYPE 4: TODAY'S MOST SUSPICIOUS
-            # =====================================================
             
-            with gr.Tab("⚠️ Most Suspicious Today"):
-                gr.Markdown("### Automatically find today's most suspicious vehicle")
-                
-                suspicious_btn = gr.Button("🔎 Find Most Suspicious", variant="primary", scale=2)
-                
-                with gr.Row():
-                    suspicious_narrative = gr.Textbox(
-                        label="📋 Investigation Summary",
-                        lines=8,
-                        interactive=False
-                    )
-                    
-                    suspicious_findings = gr.Textbox(
-                        label="🚨 Alert Details",
-                        lines=8,
-                        interactive=False
-                    )
-                
-                suspicious_analysis = gr.JSON(
-                    label="📊 Detection Analysis"
+            with gr.Tab("📊 Data Analysis"):
+                analysis_output = gr.JSON(
+                    label="Detailed Metrics & Statistics"
                 )
-                
-                suspicious_recommendations = gr.Textbox(
-                    label="✅ Action Items",
-                    lines=6,
-                    interactive=False
+        
+        investigate_btn.click(
+            fn=ask_question_wrapper,
+            inputs=[question_input],
+            outputs=[response_output, findings_output, analysis_output, gr.State(value="")]
+        )
+        
+        # Example questions
+        gr.Markdown("""
+        ### 💡 Quick Investigation Starters
+        Click any example to start exploring:
+        """)
+        
+        example_questions = [
+            "What vehicles were in Adyar at midnight last night?",
+            "Show me suspicious midnight activity in industrial zones",
+            "Which plates appear most frequently in different locations?",
+            "Analyze activity patterns in Nungambakkam",
+            "Find vehicles with multiple detections in one day",
+        ]
+        
+        with gr.Row():
+            for example in example_questions[:3]:
+                gr.Button(example, size="sm", variant="secondary").click(
+                    fn=lambda q=example: ask_question_wrapper(q),
+                    outputs=[response_output, findings_output, analysis_output]
                 )
-                
-                suspicious_btn.click(
-                    fn=find_suspicious_wrapper,
-                    inputs=[],
-                    outputs=[suspicious_narrative, suspicious_findings, suspicious_analysis, suspicious_recommendations]
+        
+        with gr.Row():
+            for example in example_questions[3:]:
+                gr.Button(example, size="sm", variant="secondary").click(
+                    fn=lambda q=example: ask_question_wrapper(q),
+                    outputs=[response_output, findings_output, analysis_output]
                 )
 
     # =====================================================