cmc

rag.py

@@ -1,3 +1,4 @@
+# rag.py – bullet-proof and container-safe
 from __future__ import annotations
 import os, re, json, requests
 from functools import lru_cache
@@ -21,6 +22,13 @@ SUPABASE_URL = os.getenv("SUPABASE_URL")
 SUPABASE_KEY = os.getenv("SUPABASE_KEY")
 HF_TOKEN     = os.getenv("HF_TOKEN")
 
+# Safe cache path for HuggingFace
+CACHE_DIR = os.getenv("HF_HOME", "/tmp/hf_cache")
+os.makedirs(CACHE_DIR, exist_ok=True)
+os.environ["TRANSFORMERS_CACHE"] = CACHE_DIR
+os.environ["HF_HOME"] = CACHE_DIR
+os.environ["HF_HUB_CACHE"] = CACHE_DIR
+
 supabase = create_client(SUPABASE_URL, SUPABASE_KEY)
 
 # ------------------------------------------------------------------ INTENT
@@ -93,6 +101,7 @@ def get_texts() -> List[str]:
 @lru_cache(maxsize=1)
 def get_vectorstore() -> FAISS:
     texts = get_texts()
+    embeddings = None
 
     try:
         from langchain_huggingface import HuggingFaceInferenceAPIEmbeddings
@@ -101,14 +110,18 @@ def get_vectorstore() -> FAISS:
             model_name="sentence-transformers/all-MiniLM-L6-v2"
         )
     except ImportError:
-        from langchain.embeddings import HuggingFaceEmbeddings
-        embeddings = HuggingFaceEmbeddings(
-            model_name="sentence-transformers/all-MiniLM-L6-v2"
-        )
-        print("⚠️ Falling back to HuggingFaceEmbeddings (local) – inference API not available")
+        try:
+            from langchain_huggingface import HuggingFaceEmbeddings
+            embeddings = HuggingFaceEmbeddings(
+                model_name="sentence-transformers/all-MiniLM-L6-v2",
+                cache_folder=CACHE_DIR
+            )
+            print("⚙️ Using HuggingFaceEmbeddings fallback")
+        except Exception as e:
+            raise RuntimeError(f"❌ Failed to load embeddings: {e}")
 
     if not texts:
-        return FAISS.from_texts([""], embeddings)
+        return FAISS.from_texts([""], embeddings)  # dummy FAISS
 
     splitter = RecursiveCharacterTextSplitter(chunk_size=600, chunk_overlap=50)
     docs = splitter.create_documents(texts, metadatas=[{"source": DATASET}] * len(texts))
@@ -124,11 +137,13 @@ def get_llm():
         huggingfacehub_api_token=HF_TOKEN
     )
 
-PROMPT = PromptTemplate.from_template("""You are Amina, assistant for {company}.
+PROMPT = PromptTemplate.from_template("""
+You are Amina, assistant for {company}.
 Use only the context below. If unsure, say: "A human agent will follow up."
 Context: {context}
 Question: {question}
-Answer:""")
+Answer:
+""")
 
 # ------------------------------------------------------------------ MAIN
 def ask_question(phone: str, question: str) -> Tuple[str, List]:
@@ -160,4 +175,4 @@ def ask_question(phone: str, question: str) -> Tuple[str, List]:
 
 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()
+    supabase.table("chat_memory").insert({"user_phone": phone, "role": "assistant", "message": a}).execute()