Enhance intent extraction logic in AI chat

- Updated the `extract_and_update_intent` function to provide clearer instructions for handling the "must_have" field, specifying how to manage additions, changes, and clarifications of requirements.
- Improved logging to reflect the LLM's decision-making process regarding the "must_have" field, enhancing traceability during intent updates.
- Added detailed context for the current intent state, improving the clarity of intent extraction and user interaction.

ai_chat.py

@@ -147,9 +147,20 @@ async def extract_and_update_intent(state):
     Extract and normalize the user's current property search intent from this message:
     {user_message}
 
-    Normalize abbreviations (e.g., 'sqm' → 'square metres') and correct any spelling mistakes.
-    If the user is asking a definition or clarification (e.g. 'What does square metre mean?'), answer that question fully and do not update the intent. After answering, wait for the next user message to extract values.
-    Return only a JSON object with keys {intent_fields}, using null for unknown.
+    Current intent state:
+    - Location: {intent.get('location_preference', 'Not set')}
+    - Budget: {intent.get('budget', 'Not set')}
+    - Size: {intent.get('size_preference_sqm', 'Not set')} sqm
+    - Must-haves: {intent.get('must_have', [])}
+
+    Instructions:
+    1. Normalize abbreviations (e.g., 'sqm' → 'square metres') and correct any spelling mistakes.
+    2. For must_have field: Determine if the user is ADDING new requirements, CHANGING their mind, or CLARIFYING existing ones.
+       - If adding: Include both existing and new items in the array
+       - If changing: Replace with new requirements
+       - If clarifying: Update with more specific versions
+    3. If the user is asking a definition or clarification (e.g. 'What does square metre mean?'), answer that question fully and do not update the intent.
+    4. Return only a JSON object with keys {intent_fields}, using null for unknown.
     """
 
     response = await llm.ainvoke([{"role":"user","content":extraction_prompt}])
@@ -195,11 +206,10 @@ async def extract_and_update_intent(state):
         if new_val is not None and new_val != old_val:
             print(f"Updating intent field {field}: {old_val} -> {new_val}")
             
-            # Handle must_have field as array
+            # Handle must_have field as array (LLM decides the logic)
             if field == "must_have" and new_val:
-                # Convert string to array format for database
+                # Convert to array format for database
                 if isinstance(new_val, str):
-                    # Split by comma if multiple items, otherwise single item
                     if "," in new_val:
                         must_have_array = [item.strip() for item in new_val.split(",")]
                     else:
@@ -207,7 +217,7 @@ async def extract_and_update_intent(state):
                 else:
                     must_have_array = new_val if isinstance(new_val, list) else [str(new_val)]
                 
-                print(f"Converting must_have to array: {must_have_array}")
+                print(f"LLM decided must_have: {must_have_array}")
                 await update_user_intent(session_id, {field: must_have_array})
                 intent[field] = must_have_array
             else: