import os from langchain_google_genai import ChatGoogleGenerativeAI, GoogleGenerativeAIEmbeddings from langchain_chroma import Chroma from src.retrieval.hybrid_retriever import get_hybrid_retriever from src.agents.final_agent import generate_answer # Moved to Top # --- GLOBAL SINGLETONS --- LLM = ChatGoogleGenerativeAI(model="gemini-3-flash-preview", temperature=0) EMBEDDINGS = GoogleGenerativeAIEmbeddings(model="models/gemini-embedding-2-preview") VECTOR_DB = Chroma(persist_directory="./chroma_db", embedding_function=EMBEDDINGS) def run_with_cache_and_eval(query): # --- 2. FAST CACHE CHECK --- cache_results = VECTOR_DB.similarity_search_with_relevance_scores( query, k=1, filter={"type": "cache"} ) if cache_results and cache_results[0][1] > 0.95: return cache_results[0][0].metadata.get("response"), 0, "YES (Cached)", [] # --- 3. LIGHTWEIGHT RETRIEVAL --- # We skip the heavy local CrossEncoder entirely retriever = get_hybrid_retriever() docs = retriever.invoke(query)[:5] # Just take top 5 # --- 4. SINGLE-PASS GENERATION --- from src.agents.final_agent import generate_answer # Ensure this is a clean call response_text = generate_answer(query, docs) # --- 5. BACKGROUND/METADATA CACHE --- VECTOR_DB.add_texts( texts=[query], metadatas=[{"response": response_text, "type": "cache"}] ) return response_text, 100, "Verified", docs