cmc

rag.py

@@ -1,6 +1,6 @@
 # rag.py – bullet-proof and container-safe
 from __future__ import annotations
-import os, re, json, requests
+import os, re, json, requests, traceback
 from functools import lru_cache
 from typing import List, Tuple
 
@@ -18,11 +18,12 @@ CONFIG      = "default"
 SPLIT       = "train"
 LIMIT       = 500
 LLM_MODEL   = "microsoft/DialoGPT-medium"
+
 SUPABASE_URL = os.getenv("SUPABASE_URL")
 SUPABASE_KEY = os.getenv("SUPABASE_KEY")
 HF_TOKEN     = os.getenv("HF_TOKEN")
 
-# Safe cache path for HuggingFace
+# ------------------------------------------------------------------ SAFE CACHE
 CACHE_DIR = os.getenv("HF_HOME", "/tmp/hf_cache")
 os.makedirs(CACHE_DIR, exist_ok=True)
 os.environ["TRANSFORMERS_CACHE"] = CACHE_DIR
@@ -31,7 +32,7 @@ os.environ["HF_HUB_CACHE"] = CACHE_DIR
 
 supabase = create_client(SUPABASE_URL, SUPABASE_KEY)
 
-# ------------------------------------------------------------------ INTENT
+# ------------------------------------------------------------------ INTENTS
 GREETING_RE  = re.compile(r"\b(hi|hello|hey|good morning|good afternoon|good evening)\b", re.I)
 THANKS_RE    = re.compile(r"\b(thank|thanks|appreciate)\b", re.I)
 BYE_RE       = re.compile(r"\b(bye|goodbye|see you|later)\b", re.I)
@@ -104,27 +105,33 @@ def get_vectorstore() -> FAISS:
     embeddings = None
 
     try:
+        # Primary attempt: use Inference API embeddings
         from langchain_huggingface import HuggingFaceInferenceAPIEmbeddings
         embeddings = HuggingFaceInferenceAPIEmbeddings(
             api_key=HF_TOKEN,
             model_name="sentence-transformers/all-MiniLM-L6-v2"
         )
+        print("✅ Using HuggingFaceInferenceAPIEmbeddings")
     except ImportError:
+        # Fallback: local embeddings (container-safe)
         try:
             from langchain_huggingface import HuggingFaceEmbeddings
             embeddings = HuggingFaceEmbeddings(
                 model_name="sentence-transformers/all-MiniLM-L6-v2",
                 cache_folder=CACHE_DIR
             )
-            print("⚙️ Using HuggingFaceEmbeddings fallback")
+            print("⚙️ Using local HuggingFaceEmbeddings fallback")
         except Exception as e:
+            traceback.print_exc()
             raise RuntimeError(f"❌ Failed to load embeddings: {e}")
 
     if not texts:
-        return FAISS.from_texts([""], embeddings)  # dummy FAISS
+        print("⚠ No dataset texts available; creating dummy FAISS index.")
+        return FAISS.from_texts([""], embeddings)
 
     splitter = RecursiveCharacterTextSplitter(chunk_size=600, chunk_overlap=50)
     docs = splitter.create_documents(texts, metadatas=[{"source": DATASET}] * len(texts))
+    print(f"🔹 Building FAISS index with {len(docs)} chunks")
     return FAISS.from_documents(docs, embeddings)
 
 # ------------------------------------------------------------------ LLM
@@ -145,7 +152,7 @@ Question: {question}
 Answer:
 """)
 
-# ------------------------------------------------------------------ MAIN
+# ------------------------------------------------------------------ MAIN LOGIC
 def ask_question(phone: str, question: str) -> Tuple[str, List]:
     intent  = _detect_intent(question)
     company = _company_from_text(question)
@@ -155,24 +162,40 @@ def ask_question(phone: str, question: str) -> Tuple[str, List]:
         _save_chat(phone, question, answer)
         return answer, []
 
-    vs   = get_vectorstore()
-    docs = vs.similarity_search(question, k=3)
+    try:
+        vs = get_vectorstore()
+        docs = vs.similarity_search(question, k=3)
+    except Exception as e:
+        print(f"❌ Vector store retrieval failed: {e}")
+        docs = []
+
     if not docs or docs[0].page_content.strip() == "":
         answer = _fallback_answer(company, intent if intent in ("money", "complain") else "default")
         _save_chat(phone, question, answer)
         return answer, []
 
-    qa = RetrievalQA.from_chain_type(
-        llm=get_llm(),
-        retriever=vs.as_retriever(search_kwargs={"k": 3}),
-        return_source_documents=True,
-        chain_type_kwargs={"prompt": PROMPT}
-    )
-    result = qa({"query": question, "company": company})
-    answer = result["result"].strip()
+    try:
+        qa = RetrievalQA.from_chain_type(
+            llm=get_llm(),
+            retriever=vs.as_retriever(search_kwargs={"k": 3}),
+            return_source_documents=True,
+            chain_type_kwargs={"prompt": PROMPT}
+        )
+        result = qa({"query": question, "company": company})
+        answer = result.get("result", "").strip() or _fallback_answer(company, "default")
+        docs_used = result.get("source_documents", [])
+    except Exception as e:
+        traceback.print_exc()
+        answer = _fallback_answer(company, "default")
+        docs_used = []
+
     _save_chat(phone, question, answer)
-    return answer, result.get("source_documents", [])
+    return answer, docs_used
 
+# ------------------------------------------------------------------ SUPABASE LOGGING
 def _save_chat(phone: str, q: str, a: str) -> None:
-    supabase.table("chat_memory").insert({"user_phone": phone, "role": "user", "message": q}).execute()
-    supabase.table("chat_memory").insert({"user_phone": phone, "role": "assistant", "message": a}).execute()
+    try:
+        supabase.table("chat_memory").insert({"user_phone": phone, "role": "user", "message": q}).execute()
+        supabase.table("chat_memory").insert({"user_phone": phone, "role": "assistant", "message": a}).execute()
+    except Exception as e:
+        print(f"⚠ Chat log save failed: {e}")