Update src/app.py

src/app.py

@@ -1,25 +1,26 @@
 import os
 import pickle
 import streamlit as st
+from typing import List
 from dotenv import load_dotenv
+
+# LangChain Imports
 from langchain_google_genai import GoogleGenerativeAIEmbeddings
 from langchain_groq import ChatGroq
 from langchain_pinecone import PineconeVectorStore
 from langchain_core.prompts import PromptTemplate
 from langchain.chains import RetrievalQA
 from langchain_core.documents import Document
+from langchain_core.retrievers import BaseRetriever
+from langchain_core.callbacks import CallbackManagerForRetrieverRun
 
 load_dotenv()
 
-# --- CONFIGURATION ---
 # --- CONFIGURATION ---
 INDEX_NAME = "branham-index"
 
 # GET ABSOLUTE PATH TO THE FILE
-# This finds the folder where app.py lives (the 'src' folder)
 BASE_DIR = os.path.dirname(os.path.abspath(__file__))
-
-# Builds the full path: .../7th_handle/src/sermon_chunks.pkl
 CHUNKS_FILE = os.path.join(BASE_DIR, "sermon_chunks.pkl")
 
 # Verify it exists immediately
@@ -28,12 +29,12 @@ if not os.path.exists(CHUNKS_FILE):
 else:
     print(f"✅ SUCCESS: Pickle file found at: {CHUNKS_FILE}")
 
-# --- SEARCH ENGINE (The "Ctrl+F" Logic) ---
+# --- SEARCH ENGINE (The "Ctrl+F" Logic for Search Mode) ---
 def search_archives(query):
     status_log = []
     results = []
     
-    # PHASE 1: LOCAL TEXT SCAN (The Logic that worked in your Test)
+    # PHASE 1: LOCAL TEXT SCAN
     if os.path.exists(CHUNKS_FILE):
         try:
             with open(CHUNKS_FILE, "rb") as f:
@@ -41,21 +42,17 @@ def search_archives(query):
             
             status_log.append(f"📂 Loaded {len(chunks)} local chunks.")
             
-            # BRUTE FORCE SCAN
             query_lower = query.lower().strip()
             count = 0
             
-            # Simple, raw loop - exactly like your test script
             for doc in chunks:
                 if query_lower in doc.page_content.lower():
                     results.append(doc)
                     count += 1
-                    if count >= 20: break # Limit to 20 matches
+                    if count >= 20: break 
             
             if results:
                 status_log.append(f"✅ FOUND {len(results)} EXACT MATCHES LOCALLY.")
-                # If we find exact matches, we return them and STOP.
-                # We do not let Pinecone pollute the results.
                 return results, status_log
             else:
                 status_log.append("⚠️ No exact matches found locally.")
@@ -65,7 +62,7 @@ def search_archives(query):
     else:
         status_log.append("❌ Pickle file missing. Skipping local search.")
 
-    # PHASE 2: PINECONE FALLBACK (Only if Phase 1 failed)
+    # PHASE 2: PINECONE FALLBACK
     status_log.append("☁️ Attempting Vector Search (Pinecone)...")
     try:
         pinecone_key = os.environ.get("PINECONE_API_KEY") or st.secrets.get("PINECONE_API_KEY")
@@ -77,47 +74,44 @@ def search_archives(query):
         embeddings = GoogleGenerativeAIEmbeddings(model="models/text-embedding-004")
         vector_store = PineconeVectorStore(index_name=INDEX_NAME, embedding=embeddings)
         
-        # STRICT THRESHOLD: 0.80
-        # If it's not 80% similar, we prefer to show nothing.
+        # STRICT THRESHOLD: 0.75
         retriever = vector_store.as_retriever(
             search_type="similarity_score_threshold",
-            search_kwargs={"k": 10, "score_threshold": 0.80} 
+            search_kwargs={"k": 10, "score_threshold": 0.75} 
         )
         
         docs = retriever.invoke(query)
         if not docs:
-            status_log.append("msg: No relevant results found in Cloud (Threshold 0.80).")
+            status_log.append("msg: No relevant results found in Cloud (Threshold 0.75).")
         return docs, status_log
 
     except Exception as e:
         status_log.append(f"❌ Cloud Error: {e}")
         return [], status_log
 
-# --- RAG CHAIN (Chat) ---
 # --- RAG CHAIN (The Chat Tool) ---
 def get_rag_chain():
-    # 1. DEFINE A SMART RETRIEVER
-    # This retriever tries Local First, then Cloud.
-    class SmartRetriever:
-        def invoke(self, query):
+    
+    # 1. DEFINE SMART RETRIEVER (Must inherit from BaseRetriever)
+    class SmartRetriever(BaseRetriever):
+        def _get_relevant_documents(
+            self, query: str, *, run_manager: CallbackManagerForRetrieverRun = None
+        ) -> List[Document]:
             print(f"🧠 Chat is thinking about: '{query}'")
             
-            # PHASE A: LOCAL LOOKUP (For specific names/places)
+            # PHASE A: LOCAL LOOKUP (Precision)
             if os.path.exists(CHUNKS_FILE):
                 try:
                     with open(CHUNKS_FILE, "rb") as f:
                         chunks = pickle.load(f)
                     
-                    # Extract potential keywords (simple split)
                     keywords = [w for w in query.split() if len(w) > 3]
                     local_matches = []
                     
-                    # Scan for matches
                     for doc in chunks:
-                        # If any significant keyword from the question is in the text
                         if any(k.lower() in doc.page_content.lower() for k in keywords):
                             local_matches.append(doc)
-                            if len(local_matches) >= 15: break # Cap at 15 to fit in AI memory
+                            if len(local_matches) >= 15: break
                     
                     if local_matches:
                         print(f"✅ Chat found {len(local_matches)} local clues.")
@@ -125,9 +119,14 @@ def get_rag_chain():
                 except:
                     pass
             
-            # PHASE B: CLOUD FALLBACK (For general concepts)
+            # PHASE B: CLOUD FALLBACK (Concepts)
             print("☁️ Checking Cloud...")
             try:
+                pinecone_key = os.environ.get("PINECONE_API_KEY") or st.secrets.get("PINECONE_API_KEY")
+                google_key = os.environ.get("GOOGLE_API_KEY") or st.secrets.get("GOOGLE_API_KEY")
+                os.environ["PINECONE_API_KEY"] = pinecone_key
+                os.environ["GOOGLE_API_KEY"] = google_key
+
                 embeddings = GoogleGenerativeAIEmbeddings(model="models/text-embedding-004")
                 vector_store = PineconeVectorStore(index_name=INDEX_NAME, embedding=embeddings)
                 retriever = vector_store.as_retriever(search_kwargs={"k": 10})
@@ -136,20 +135,20 @@ def get_rag_chain():
                 print(f"❌ Cloud Error: {e}")
                 return []
 
-    # 2. SETUP THE LLM
+    # 2. SETUP LLM
     groq_key = os.environ.get("GROQ_API_KEY") or st.secrets.get("GROQ_API_KEY")
     os.environ["GROQ_API_KEY"] = groq_key
     
     llm = ChatGroq(model="llama-3.3-70b-versatile", temperature=0.3)
 
-    # 3. THE PROMPT (Brother Branham's Voice)
+    # 3. PROMPT
     template = """You are William Marion Branham.
     
     INSTRUCTIONS:
     1. Answer the question using ONLY the provided CONTEXT.
     2. If the Context contains the answer, quote it and explain it simply.
     3. If the Context is empty or irrelevant, say: "Brother, I do not find that specific record here."
-    4. Speak with humility (e.g., "The Lord showed me," "I said").
+    4. Speak with humility.
 
     CONTEXT:
     {context}
@@ -160,11 +159,11 @@ def get_rag_chain():
     
     PROMPT = PromptTemplate(template=template, input_variables=["context", "question"])
     
-    # 4. BUILD THE CHAIN
+    # 4. BUILD CHAIN
     chain = RetrievalQA.from_chain_type(
         llm=llm,
         chain_type="stuff",
-        retriever=SmartRetriever(), # <--- We use our custom smart logic
+        retriever=SmartRetriever(), # Now a valid BaseRetriever
         return_source_documents=True,
         chain_type_kwargs={"prompt": PROMPT, "document_variable_name": "context"},
         input_key="question"