Update chatbot_rag.py

chatbot_rag.py

@@ -8,18 +8,23 @@ from langchain.chains import RetrievalQA
 
 def build_qa():
     """Builds and returns the RAG QA pipeline."""
+    print("🚀 Starting QA pipeline...")
 
     # 1. Embeddings
+    print("🔹 Loading embeddings...")
     embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")
 
-    # 2. Load vector DB (must already exist in ./db)
+    # 2. Load vector DB
+    print("🔹 Loading Chroma DB...")
     vectorstore = Chroma(
         persist_directory="db",
         collection_name="rag-docs",
         embedding_function=embeddings,
     )
+    print("📂 Docs in DB:", vectorstore._collection.count())
 
-    # 3. LLM (lighter model = faster inference)
+    # 3. LLM
+    print("🔹 Loading LLM...")
     model_id = "microsoft/phi-3-mini-4k-instruct"
     tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
     model = AutoModelForCausalLM.from_pretrained(
@@ -27,6 +32,7 @@ def build_qa():
         device_map="auto",
         torch_dtype="auto"
     )
+    print("✅ LLM loaded.")
 
     pipe = pipeline(
         "text-generation",
@@ -35,10 +41,10 @@ def build_qa():
         max_new_tokens=256,
         temperature=0.2,
     )
-
     llm = HuggingFacePipeline(pipeline=pipe)
 
-    # 4. RAG chain
+    # 4. QA Chain
+    print("🔹 Building RetrievalQA...")
     retriever = vectorstore.as_retriever(search_kwargs={"k": 3})
     qa = RetrievalQA.from_chain_type(
         llm=llm,
@@ -46,19 +52,5 @@ def build_qa():
         return_source_documents=False
     )
 
+    print("✅ QA pipeline ready.")
     return qa
-
-
-# Build at import time (so it's ready when app runs)
-try:
-    qa_pipeline = build_qa()
-except Exception as e:
-    qa_pipeline = None
-    print("❌ Failed to build QA pipeline:", e)
-
-
-def get_answer(query: str) -> str:
-    """Takes user query and returns chatbot response."""
-    if qa_pipeline is None:
-        return "⚠️ QA pipeline not initialized."
-    return qa_pipeline.run(query)