Lab 4: Hybrid RAG ChatBot

Difficulty: Intermediate · Estimated time: ~4 hours

Objective

Extend your Lab 2 RAG system with:

1. PGVector for persistent vector storage
2. BM25 + dense hybrid search with fusion
3. Metadata filtering on document properties
4. Multi-query retrieval for better recall
5. Streamlit UI for interactive chat

Step 1 — Setup PGVector

python

import psycopg2
from pgvector.psycopg2 import register_vector

conn = psycopg2.connect("postgresql://user:pass@localhost:5432/tds_rag")
register_vector(conn)
conn.execute("""
    CREATE TABLE IF NOT EXISTS documents (
        id SERIAL PRIMARY KEY,
        content TEXT,
        embedding VECTOR(1536),
        metadata JSONB
    )
""")

Step 2 — Multi-query retrieval

python

def multi_query_retrieve(question: str, num_queries: int = 3):
    """Generate multiple query variations for better recall."""
    variations = generate_query_variations(question, num_queries)
    all_results = []
    for q in variations:
        results = hybrid_search(q, top_k=5)
        all_results.extend(results)
    return deduplicate_and_rerank(all_results)

Step 3 — Streamlit UI

python

import streamlit as st

st.title("TDS Hybrid RAG ChatBot")
if "messages" not in st.session_state:
    st.session_state.messages = []

for msg in st.session_state.messages:
    st.chat_message(msg["role"]).write(msg["content"])

if prompt := st.chat_input("Ask about the course..."):
    st.session_state.messages.append({"role": "user", "content": prompt})
    with st.chat_message("assistant"):
        response = rag_pipeline(prompt)
        st.write(response)
        st.session_state.messages.append({"role": "assistant", "content": response})

Submission

GitHub repo + Docker Compose setup with PostgreSQL + Streamlit app

Grading rubric

Criterion	Points
PGVector stores and retrieves vectors	20
Hybrid search with BM25 + dense fusion	25
Metadata filtering works correctly	15
Multi-query improves retrieval quality	15
Streamlit UI is functional	25
Total	100