Update src/vectorstore.py

src/vectorstore.py

@@ -2,34 +2,36 @@ import faiss
 import numpy as np
 
 # -----------------------------
-# BUILD FAISS INDEX
+# BUILD FAISS INDEX (Optimized for Cosine Similarity)
 # -----------------------------
 def build_faiss_index(embeddings: list):
     """
-    📘 Builds a FAISS index for similarity search.
+    📘 Builds a FAISS index optimized for cosine similarity (fast + accurate).
 
     Args:
         embeddings (list): List of embedding vectors (lists of floats).
     Returns:
-        faiss.IndexFlatL2: FAISS index for vector similarity.
+        faiss.IndexFlatIP: FAISS index for cosine similarity search.
     """
     if not embeddings or not isinstance(embeddings, list):
         raise ValueError("❌ Invalid input: embeddings must be a non-empty list.")
 
     # Convert to numpy float32 array
     vectors = np.array(embeddings).astype("float32")
-    dim = vectors.shape[1]  # Get embedding dimension (e.g., 384 or 768)
+    dim = vectors.shape[1]
 
-    # Create index using Euclidean (L2) distance
-    index = faiss.IndexFlatL2(dim)
+    # ✅ Normalize embeddings (turns dot product into cosine similarity)
+    faiss.normalize_L2(vectors)
 
-    # Add vectors to index
+    # ✅ Use Inner Product index (fast cosine similarity)
+    index = faiss.IndexFlatIP(dim)
     index.add(vectors)
-    print(f"✅ FAISS index built with {index.ntotal} vectors of dimension {dim}")
 
+    print(f"✅ FAISS index built with {index.ntotal} vectors of dimension {dim} (cosine similarity)")
     return index
 
 
+
 # -----------------------------
 # SEARCH FAISS INDEX
 # -----------------------------