Update main.py

main.py

@@ -1,10 +1,10 @@
 """
-main.py — Pricelyst Shopping Advisor (Jessica Edition 2026 - Upgrade v3.0)
+main.py — Pricelyst Shopping Advisor (Jessica Edition 2026 - Upgrade v3.1)
 
-✅ Feature: "Concept Exploder" (Converts "Plan a Braai" -> Shopping List).
-✅ Feature: "Hybrid Valuation" (Estimates prices for missing items in Plans).
-✅ Feature: "Market Intelligence" (Pre-calculated Voice Context).
-✅ UI Match: Restored v1 Markdown Tables & Creative Tips.
+✅ Feature: "Vernacular Engine" (Shona/Ndebele/English Input -> Native Response).
+✅ Feature: "Precision Search" (Prioritizes exact phrase matches over popularity).
+✅ Feature: "Concept Exploder" (Event Planning -> Shopping List).
+✅ UI/UX: "Nearest Match" phrasing for substitutions.
 ✅ Core: Deep Vector Search + Market Matrix + Store Preferences.
 
 ENV VARS:
@@ -236,39 +236,60 @@ def get_market_index(force_refresh: bool = False) -> pd.DataFrame:
     return _data_cache["df"]
 
 # =========================
-# 2. Analyst Engine (Matrix & Calculations)
+# 2. Analyst Engine (Precision Search & Matrix)
 # =========================
 
 def search_products_deep(df: pd.DataFrame, query: str, limit: int = 15) -> pd.DataFrame:
+    """
+    Precision Search Algorithm.
+    Prioritizes:
+    1. Exact sequential match in Name/Vector (Highest Score)
+    2. Token overlap (Medium Score)
+    3. Views/Popularity (Tie-breaker)
+    """
     if df.empty or not query: return df
     q_norm = _norm(query)
+    q_tokens = set(q_norm.split())
     
-    # 1. Exact/Partial Vector Match
-    mask = df['search_vector'].str.contains(q_norm, regex=False)
-    matches = df[mask].copy()
-    
-    # 2. Token Overlap Fallback
-    if matches.empty:
-        q_tokens = set(q_norm.split())
-        def token_score(text):
-            if not isinstance(text, str): return 0
-            text_tokens = set(text.split())
-            if not text_tokens: return 0
-            return len(q_tokens.intersection(text_tokens))
+    def scoring_algo(row):
+        score = 0
+        vector = row['search_vector']
+        
+        # 1. Exact Name Match (Highest)
+        if q_norm == _norm(row['product_name']):
+            score += 1000
+            
+        # 2. Sequential Vector Match (High)
+        if q_norm in vector:
+            score += 500
+            
+        # 3. Brand Match
+        if row['brand'].lower() in q_norm:
+            score += 200
+            
+        # 4. Token Overlap
+        text_tokens = set(vector.split())
+        overlap = len(q_tokens.intersection(text_tokens))
+        score += (overlap * 50)
         
-        df_scored = df.copy()
-        df_scored['score'] = df_scored['search_vector'].apply(token_score)
-        matches = df_scored[df_scored['score'] > 0]
+        return score
 
+    df_scored = df.copy()
+    df_scored['match_score'] = df_scored.apply(scoring_algo, axis=1)
+    
+    # Filter out zero matches
+    matches = df_scored[df_scored['match_score'] > 0]
+    
     if matches.empty: return matches
 
-    matches = matches.sort_values(by=['views', 'price'], ascending=[False, True])
+    # Sort: Match Score (Desc) -> Views (Desc) -> Price (Asc)
+    matches = matches.sort_values(by=['match_score', 'views', 'price'], ascending=[False, False, True])
+    
     return matches.head(limit)
 
 def calculate_basket_optimization(item_names: List[str], preferred_retailer: str = None) -> Dict[str, Any]:
     """
-    Generates a FULL MARKET MATRIX.
-    Returns best store, plus how EVERY other store performed.
+    Generates a FULL MARKET MATRIX with Precision Search.
     """
     df = get_market_index()
     if df.empty: 
@@ -291,11 +312,13 @@ def calculate_basket_optimization(item_names: List[str], preferred_retailer: str
         q_norm = _norm(item)
         res_norm = _norm(best_match['product_name'] + " " + best_match['brand'])
         q_tokens = q_norm.split()
+        
         is_substitute = False
-        if len(q_tokens) > 1:
-            found_tokens = sum(1 for t in q_tokens if t in res_norm)
-            if found_tokens < len(q_tokens) / 2: 
-                is_substitute = True
+        # If query has brand/spec but result score is low-ish (not exact name match), flag it.
+        # Using a simple heuristic for now based on token overlap vs query length
+        found_tokens = sum(1 for t in q_tokens if t in res_norm)
+        if len(q_tokens) > 1 and found_tokens < len(q_tokens):
+            is_substitute = True
                 
         # Aggregate all offers
         product_offers = hits[hits['product_name'] == best_match['product_name']].sort_values('price')
@@ -385,7 +408,7 @@ def calculate_zesa_units(amount_usd: float) -> Dict[str, Any]:
     }
 
 # =========================
-# 3. Gemini Helpers (The Intelligence)
+# 3. Gemini Helpers (Vernacular & Intelligence)
 # =========================
 
 def gemini_detect_intent(transcript: str) -> Dict[str, Any]:
@@ -401,10 +424,11 @@ def gemini_detect_intent(transcript: str) -> Dict[str, Any]:
     - EVENT_PLANNING: "Plan a braai", "Wedding list", "Dinner for 5" (Implicit lists).
     
     Extract:
-    - items: list of specific products found.
+    - items: list of specific products found. **TRANSLATE ALL ITEMS TO ENGLISH** (e.g. 'Hupfu' -> 'Maize Meal').
     - utility_amount: number
     - store_preference: if a specific store is named (e.g. "at OK Mart").
     - is_event_planning: boolean (true if user asks to plan an event but lists no items).
+    - language: Detected user language (e.g., "Shona", "Ndebele", "English").
     
     JSON Schema:
     {
@@ -413,7 +437,8 @@ def gemini_detect_intent(transcript: str) -> Dict[str, Any]:
       "items": ["string"],
       "utility_amount": number,
       "store_preference": "string",
-      "is_event_planning": boolean
+      "is_event_planning": boolean,
+      "language": "string"
     }
     """
     try:
@@ -422,10 +447,10 @@ def gemini_detect_intent(transcript: str) -> Dict[str, Any]:
             contents=PROMPT + "\nTranscript: " + transcript,
             config=types.GenerateContentConfig(response_mime_type="application/json")
         )
-        return _safe_json_loads(resp.text, {"actionable": False, "intent": "CASUAL_CHAT"})
+        return _safe_json_loads(resp.text, {"actionable": False, "intent": "CASUAL_CHAT", "language": "English"})
     except Exception as e:
         logger.error(f"Intent Detect Error: {e}")
-        return {"actionable": False, "intent": "CASUAL_CHAT"}
+        return {"actionable": False, "intent": "CASUAL_CHAT", "language": "English"}
 
 def gemini_explode_concept(transcript: str) -> List[str]:
     """
@@ -436,7 +461,7 @@ def gemini_explode_concept(transcript: str) -> List[str]:
     PROMPT = f"""
     User wants to plan an event: "{transcript}".
     Generate a STRICT list of 10-15 essential Zimbabwean shopping items for this.
-    Use local terms (e.g., 'Boerewors', 'Maize Meal', 'Mazoe', 'Charcoal').
+    Use English terms for database lookup (e.g. 'Maize Meal', 'Cooking Oil').
     Return ONLY a JSON list of strings.
     """
     try:
@@ -492,29 +517,35 @@ def gemini_chat_response(transcript: str, intent: Dict, analyst_data: Dict, chat
     if analyst_data:
         context_str += f"ANALYST DATA: {json.dumps(analyst_data, default=str)}\n"
 
+    language = intent.get("language", "English")
+
     PROMPT = f"""
     You are Jessica, Pricelyst's Shopping Advisor (Zimbabwe).
     Role: Intelligent Shopping Companion.
     Goal: Shortest path to value. Complete Transparency.
     
     INPUT: "{transcript}"
+    USER LANGUAGE: {language}
     INTENT: {intent.get('intent')}
     CONTEXT:
     {context_str}
     
     LOGIC RULES:
     
-    1. **BASKET COMPARISON**:
+    1. **LANGUAGE**: Reply in **{language}**. If Shona, use Shona. If English, use English.
+    
+    2. **BASKET COMPARISON**:
        - If `market_matrix` has multiple stores, compare totals.
        - "Spar is **$6.95**, OK Mart is **$4.00** (but missing Oil)."
 
-    2. **BRAND SUBSTITUTES**:
-       - If `is_substitute` is TRUE: "I couldn't find **[Query]**, so I used **[Found]** ($Price) as a placeholder."
+    3. **BRAND SUBSTITUTES (Phrasing)**:
+       - If `is_substitute` is TRUE for an item, say: 
+         "I couldn't find **[Query]**, but the **nearest match is** **[Found]** ($Price)."
 
-    3. **SINGLE ITEMS**:
+    4. **SINGLE ITEMS**:
        - Best price first, then others.
 
-    4. **CASUAL**:
+    5. **CASUAL**:
        - Reset if user says "Hi".
        
     TONE: Helpful, direct, Zimbabwean. Use Markdown.
@@ -579,7 +610,7 @@ def health():
         "ok": True,
         "offers_indexed": len(df),
         "api_source": PRICE_API_BASE,
-        "persona": "Jessica v3.0 (Event Planner)"
+        "persona": "Jessica v3.1 (Babel Fish)"
     })
 
 @app.post("/chat")
@@ -747,7 +778,7 @@ def call_briefing():
 def log_call_usage():
     """
     Post-Call Orchestrator.
-    v3.0 Upgrade: Handles Concept Explosion for Event Planning.
+    v3.1: Handles Concept Explosion & Plan Generation.
     """
     body = request.get_json(silent=True) or {}
     pid = body.get("profile_id")
@@ -780,7 +811,7 @@ def log_call_usage():
         if target_items:
             analyst_result = calculate_basket_optimization(target_items)
             
-            # v3.0: Even if missing items, we generate plan because prompt will ESTIMATE them
+            # v3.1: Generate Plan with Estimates & Creative Tips
             md_content = gemini_generate_4step_plan(transcript, analyst_result)
             
             plan_data = {