Intermediate55 minProduction RAG

Embeddings & Vector Search

Embed documents, store in pgvector, and implement cosine + ANN search with metadata filtering.

Scenario

Build the retrieval layer: embed documents, store in pgvector, and query with metadata filters for access control.

Objective

Implement a VectorStore with embed, insert, and search(filter) methods.

Starter code

Implement the TODOs to complete the lab.

vectorstore.pypython

import pgvector
from openai import OpenAI

client = OpenAI()

class VectorStore:
    def embed(self, text: str) -> list[float]:
        """Embed text using text-embedding-3-small."""
        # TODO
        pass

    def insert(self, text: str, metadata: dict) -> str:
        """Insert document with embedding and metadata."""
        # TODO
        pass

    def search(self, query: str, k: int = 5, filters: dict = None) -> list[dict]:
        """Search by similarity with optional metadata filter."""
        # TODO — filter must apply at SQL level, not display
        pass

Solution hints

1Use pgvector's <=> operator for cosine distance
2Apply metadata filters in the WHERE clause
3Create an HNSW index for ANN
4Return chunk_id, score, text, metadata

Validation steps

Your implementation should pass these checks.

Insert 10 docs, search returns relevant ones
Metadata filter excludes docs not matching filter
Search respects HNSW index (check EXPLAIN)
Cosine similarity scores are in [-1, 1]

Run validation

This is a simulated validation environment. In production, this would run your code against the validation steps.