Update src/services/db_client.py

src/services/db_client.py

@@ -120,17 +120,47 @@ def search_faces(idx, vec: List[float], det_score: float) -> Dict[str, Any]:
             }
     return image_map
 
-def search_objects(idx, vec: List[float]) -> List[Dict[str, Any]]:
-    res = idx.query(vector=vec, top_k=50, include_metadata=True)
+import numpy as np
+
+def search_objects(idx, vec: List[float], filter_dict: dict = None) -> List[Dict[str, Any]]:
+    query_kwargs = {"vector": vec, "top_k": 50, "include_metadata": True}
+    if filter_dict:
+        query_kwargs["filter"] = filter_dict
+        
+    res = idx.query(**query_kwargs)
+    matches = res.get("matches", [])
+    
+    if not matches:
+        return []
+
+    # ── ENTERPRISE FIX: Dynamic Gradient Analysis ──
+    # Extract the raw scores
+    scores = [m.get("score", 0) for m in matches]
+    
+    # Calculate the drop-off from the absolute best match to the 5th match
+    if len(scores) >= 5:
+        top_score = scores[0]
+        fifth_score = scores[4]
+        gradient = top_score - fifth_score
+        
+        # If the highest score is mediocre AND there is no statistical "cliff", 
+        # it means the AI just grabbed a random cluster of distant neighbors.
+        # This dynamically catches out-of-distribution items without hardcoding 
+        # strict global cutoffs.
+        if top_score < 0.65 and gradient < 0.05:
+            return [] # System realizes it's hallucinating and returns nothing
+
+    # Proceed to map results normally...
     results = []
-    for match in res.get("matches", []):
+    for match in matches:
         meta = match.get("metadata", {})
         results.append({
             "url": meta.get("url", ""),
-            "score": round(match.get("score", 0) * 100, 2),
+            "score": round(match.get("score", 0), 4),
             "raw_score": match.get("score", 0),
             "folder": meta.get("folder", "uncategorized")
         })
+        
     return results
 
 def merge_face_results(groups: List[Dict[str, Any]]) -> List[Dict[str, Any]]: