import faiss import numpy as np import pandas as pd from sentence_transformers import SentenceTransformer import re class QuoteRAG: def __init__(self, model_path="models/fine_tuned_model", data_path="english_quotes.csv"): # Load model try: self.model = SentenceTransformer(model_path) print("Loaded fine-tuned model") except: self.model = SentenceTransformer("all-MiniLM-L6-v2") print("Loaded base model") # Load dataset self.df = pd.read_csv(data_path) # Encode all quotes self.embeddings = self.model.encode(self.df["quote"].tolist(), convert_to_numpy=True) d = self.embeddings.shape[1] # Build FAISS index self.index = faiss.IndexFlatL2(d) self.index.add(self.embeddings) print("FAISS index built with", len(self.df), "quotes") def search(self, query, top_k=5): # Encode query query_emb = self.model.encode([query], convert_to_numpy=True) distances, indices = self.index.search(query_emb, top_k * 3) # fetch more for filtering results = [] for idx, dist in zip(indices[0], distances[0]): row = self.df.iloc[idx] # Normalized similarity: 0–1 (higher is better) similarity = round(1 / (1 + float(dist)), 3) results.append({ "quote": row["quote"], "author": row["author"], "tags": row.get("tags", ""), "similarity": similarity }) # Simple author filter if author name is in query query_lower = query.lower() author_filtered = [r for r in results if r["author"].lower() in query_lower] if author_filtered: results = author_filtered[:top_k] else: results = results[:top_k] return results if __name__ == "__main__": rag = QuoteRAG() query = "Quotes about insanity attributed to Einstein" results = rag.search(query) for r in results: print(r)