Skip to main content

Building a RAG Pipeline with BAML

This tutorial shows how to build a RAG (Retrieval-Augmented Generation) pipeline using BAML's structured output capabilities.

Overview​

A RAG pipeline typically involves:

  1. Retrieving relevant documents
  2. Formatting context for the LLM
  3. Generating a structured response

BAML excels at step 3 - ensuring the LLM output is properly typed and validated.

Defining the Response Structure​

class RAGResponse {
  answer string @description("Direct answer to the question")
  sources Citation[] @description("Citations from retrieved documents")
  confidence float @description("Confidence score 0-1")
}

class Citation {
  document_id string
  quote string
  relevance float
}

The RAG Function​

function AnswerWithSources(
  question: string,
  context: Document[]
) -> RAGResponse {
  client GPT4
  prompt #"
    Answer the following question using ONLY the provided context.

    Question: {{ question }}

    Context:
    {% for doc in context %}
    [Doc {{ doc.id }}]: {{ doc.content }}
    {% endfor %}

    Provide your answer with citations.
  "#
}