Update utils/fallback_suggester.py

utils/fallback_suggester.py

@@ -3,24 +3,40 @@
 import json
 from sentence_transformers import SentenceTransformer, util
 
+# 🔍 Load pre-trained semantic similarity model
 model = SentenceTransformer("sentence-transformers/paraphrase-mpnet-base-v2")
 
+# 📚 Load fallback clause database
 with open("fallback_clauses.json", "r", encoding="utf-8") as f:
     clause_bank = json.load(f)
 
+# 🔑 Extract clause labels and text
 clause_labels = list(clause_bank.keys())
 clause_texts = list(clause_bank.values())
 clause_embeddings = model.encode(clause_texts, convert_to_tensor=True)
 
-def suggest_fallback(input_clause):
+def suggest_fallback(input_clause: str, top_k: int = 3):
+    """
+    Suggest top-k fallback clauses based on semantic similarity.
+
+    Args:
+        input_clause (str): The clause to analyze.
+        top_k (int): Number of fallback suggestions to return.
+
+    Returns:
+        str: Formatted fallback suggestions.
+    """
     if not input_clause or len(input_clause.strip()) == 0:
-        return "No input clause provided."
+        return "⚠️ No input clause provided."
 
     input_embedding = model.encode(input_clause, convert_to_tensor=True)
     scores = util.cos_sim(input_embedding, clause_embeddings)[0]
-    best_idx = scores.argmax().item()
+    top_indices = scores.topk(k=min(top_k, len(clause_labels))).indices.tolist()
 
-    label = clause_labels[best_idx]
-    suggestion = clause_texts[best_idx]
+    results = []
+    for idx in top_indices:
+        label = clause_labels[idx]
+        suggestion = clause_texts[idx]
+        results.append(f"🔹 {label}:\n{suggestion}")
 
-    return f"🔹 {label} → {suggestion}"
+    return "\n\n".join(results)