Database / pgvector basics Interview Questions
How does pgvector fit into a RAG (Retrieval-Augmented Generation) pipeline?
RAG (Retrieval-Augmented Generation) is a technique that improves LLM responses by retrieving relevant documents from a knowledge base and including them as context in the prompt. pgvector serves as the vector store component, storing document embeddings and enabling semantic retrieval.
| Stage | What happens | pgvector role |
|---|---|---|
| 1. Ingest | Split documents into chunks; embed each chunk | Store chunks + embeddings in vector table |
| 2. Retrieve | Embed user query; find similar chunks | KNN query returns top-k relevant chunks |
| 3. Generate | Inject retrieved chunks into LLM prompt | No role - LLM (OpenAI, Gemini, etc.) generates answer |
import psycopg2 from pgvector.psycopg2 import register_vector from openai import OpenAI conn = psycopg2.connect("postgresql://user:pass@localhost/mydb") register_vector(conn) cur = conn.cursor() oai = OpenAI() # STAGE 1: INGEST - embed and store documents documents = [ "pgvector is a PostgreSQL extension for vector search.", "HNSW indexes provide fast approximate nearest neighbour search.", "Cosine distance is commonly used for text embeddings.", ] for text in documents: emb = oai.embeddings.create( model="text-embedding-3-small", input=text ).data[0].embedding cur.execute( "INSERT INTO documents (content, embedding) VALUES (%s, %s)", (text, emb) ) conn.commit() # STAGE 2: RETRIEVE - semantic search for user query user_question = "What kind of index should I use for fast search?" q_emb = oai.embeddings.create( model="text-embedding-3-small", input=user_question ).data[0].embedding cur.execute( "SELECT content FROM documents ORDER BY embedding <=> %s LIMIT 3", (q_emb,) ) context_chunks = [r[0] for r in cur.fetchall()] # STAGE 3: GENERATE - pass context to LLM context = "\n".join(context_chunks) completion = oai.chat.completions.create( model="gpt-4o-mini", messages=[ {"role": "system", "content": f"Answer using this context:\n{context}"}, {"role": "user", "content": user_question} ] ) print(completion.choices[0].message.content)
In a RAG pipeline, what is the role of pgvector?
What must be done to a user's query before using it to retrieve documents in a RAG pipeline?
More Related questions...
