feat: implement hybrid similarity ranking engine with adaptive weighting and score normalization

src/similarity_model/hybrid_ranker.py

@@ -124,6 +124,14 @@ def compute_hybrid_score(
         coverage_w * coverage
     )
 
+    # ==========================================
+    # Penalty for weak feature overlap 
+    # (prevents generic keyword matches from scoring too high)
+    # ==========================================
+    if feature_score < 0.40:
+        penalty = 0.50 + (feature_score / 0.40) * 0.50
+        score *= penalty
+
     return round(clamp(score), 4)
 
 

src/similarity_model/similarity_engine.py

@@ -160,9 +160,13 @@ def compare_candidate(
     feature_model
 ) -> Dict[str, Any]:
 
+    candidate_features = safe_feature_list(
+        candidate_row[FEATURE_COL]
+    )
+
     feature_result = compute_feature_similarity(
         query_project[FEATURE_COL],
-        candidate_row[FEATURE_COL],
+        candidate_features,
         model=feature_model
     )
 
@@ -235,7 +239,10 @@ def compare_candidate(
             feature_result["unique_a"],
 
         "unique_candidate_features":
-            feature_result["unique_b"]
+            feature_result["unique_b"],
+            
+        "candidate_features":
+            candidate_features
     }