Skip to content

Module 1: Vector Search

Without grounding in real data, LLMs hallucinate confidently. Vector search lets your agent retrieve actual policy documents from Redis by meaning, not just keywords, so every answer is backed by source material.

Exercise

Open exercises/digital-native/vector_search.py. You will write ~5 lines in SimpleRAGService .

create_vector_query(embedding, rag_config)

This method builds a VectorQuery — the object that tells Redis how to search your document vectors and what to send back.

Parameter What it means What to pass
vector The user's question as an embedding. Redis compares this against all stored document vectors using cosine similarity. embedding
vector_field_name Which field in your documents contains the embedding vectors. rag_config.vector_field
return_fields Which fields to return from matching documents (title, content, etc.). rag_config.return_fields
num_results How many documents to retrieve. More results = more context but weaker matches. rag_config.num_results

Return a VectorQuery(...) instance.

Show solution 
def create_vector_query(self, embedding, rag_config):
    return VectorQuery(
        vector=embedding,
        vector_field_name=rag_config.vector_field,
        return_fields=rag_config.return_fields,
        num_results=rag_config.num_results,
    )

Verify

Stop the running server (Ctrl+C in the terminal), then restart with make dev. Open localhost:3040.

  • Ask: "Why is my order running late?"
  • View the Activity Panel — you see RAG retrieval results with document titles

Try this

  • Change num_results to 1, refresh, and ask the same question. Notice how the response changes with less information.