clean final: HF Inference API embeddings (no disk)

rag.py

@@ -1,3 +1,4 @@
+# rag.py – bullet-proof: online fetch with fallback on any error
 from __future__ import annotations
 import os, re, json, requests
 from functools import lru_cache
@@ -5,7 +6,7 @@ from typing import List, Tuple
 
 from langchain.text_splitter import RecursiveCharacterTextSplitter
 from langchain_community.vectorstores import FAISS
-from langchain_huggingface import HuggingFaceEndpoint, HuggingFaceInferenceAPIEmbeddings  # ✅ FIXED: added missing import
+from langchain_huggingface import HuggingFaceEndpoint
 from langchain_core.prompts import PromptTemplate
 from langchain.chains import RetrievalQA
 from supabase import create_client
@@ -24,7 +25,6 @@ HF_TOKEN     = os.getenv("HF_TOKEN")
 supabase = create_client(SUPABASE_URL, SUPABASE_KEY)
 
 # ------------------------------------------------------------------ INTENT
-import re
 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)
@@ -94,14 +94,28 @@ def get_texts() -> List[str]:
 @lru_cache(maxsize=1)
 def get_vectorstore() -> FAISS:
     texts = get_texts()
+    embeddings = None
 
-    embeddings = HuggingFaceInferenceAPIEmbeddings(
-        api_key=HF_TOKEN,
-        model_name="sentence-transformers/all-MiniLM-L6-v2"
-    )
+    try:
+        # Try new API first
+        from langchain_huggingface import HuggingFaceInferenceAPIEmbeddings
+        embeddings = HuggingFaceInferenceAPIEmbeddings(
+            api_key=HF_TOKEN,
+            model_name="sentence-transformers/all-MiniLM-L6-v2"
+        )
+    except ImportError:
+        try:
+            # Fallback for updated version
+            from langchain_huggingface import HuggingFaceEmbeddings
+            embeddings = HuggingFaceEmbeddings(
+                model_name="sentence-transformers/all-MiniLM-L6-v2"
+            )
+            print("⚙️ Using HuggingFaceEmbeddings fallback")
+        except Exception as e:
+            raise RuntimeError(f"❌ Failed to load embeddings: {e}")
 
     if not texts:
-        return FAISS.from_texts([""], embeddings)  # dummy
+        return FAISS.from_texts([""], embeddings)  # dummy FAISS instance
 
     splitter = RecursiveCharacterTextSplitter(chunk_size=600, chunk_overlap=50)
     docs = splitter.create_documents(texts, metadatas=[{"source": DATASET}] * len(texts))
@@ -111,17 +125,19 @@ def get_vectorstore() -> FAISS:
 @lru_cache(maxsize=1)
 def get_llm():
     return HuggingFaceEndpoint(
-        repo_id="microsoft/DialoGPT-medium",
+        repo_id=LLM_MODEL,
         temperature=0.1,
         max_new_tokens=150,
         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]:
@@ -153,4 +169,4 @@ 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()