Update reasoning_engine.py

reasoning_engine.py

@@ -350,11 +350,29 @@ def retrieve_experiences(query: str, top_k: int = 3) -> List[Dict[str, Any]]:
         if not rows:
             return []
 
+        # 1. Get the embedding model from STATE
+        embed_model = STATE.get("embed_model")
+        if embed_model is None:
+            # Fallback if model isn't initialized yet
+            return []
+
+        # 2. Embed the incoming user query
         qv = _embed(query)
-        scored = []
-        for row in rows:
-            rv = _embed(row[1])
-            scored.append((float(np.dot(qv, rv)), row))
+
+        # 3. Batch embed all historical queries retrieved from the database
+        db_queries = [row[1] for row in rows]
+        rv_matrix = embed_model.encode(
+            db_queries,
+            normalize_embeddings=True,
+            convert_to_numpy=True,
+            show_progress_bar=False
+        )
+
+        # 4. Use matrix multiplication to calculate all scores simultaneously
+        scores = rv_matrix @ qv
+
+        # 5. Pair scores back up with their rows and sort
+        scored = [(float(score), row) for score, row in zip(scores, rows)]
         scored.sort(key=lambda x: x[0], reverse=True)
 
         return [
@@ -366,6 +384,7 @@ def retrieve_experiences(query: str, top_k: int = 3) -> List[Dict[str, Any]]:
                 "answer": row[4],
                 "quality": row[5],
             }
+            # Slid down to only return the top_k best matches
             for score, row in scored[:top_k]
         ]
     except Exception as exc:
@@ -539,6 +558,9 @@ Then give the final answer in a useful format.
 
 
 def plan(query: str, history: List[Dict[str, str]]) -> Dict[str, Any]:
+    # FIX: Initialize models here so retrieve_experiences has access to the embedding model
+    initialize_model()
+    
     experiences = retrieve_experiences(query, top_k=3)
     experience_text = json.dumps(experiences, ensure_ascii=False, indent=2)
 
@@ -579,7 +601,6 @@ def plan(query: str, history: List[Dict[str, str]]) -> Dict[str, Any]:
         "experiences": experiences,
     }
 
-
 def analyze(query: str, plan_data: Dict[str, Any], evidence: str, history: List[Dict[str, str]]) -> str:
     messages = [
         {"role": "system", "content": ANALYST_SYSTEM},