This lesson's interactive features are locked, please to keep using them

Augmented Generation

So far, we've been solely focused on retrieving information. Now we'll explore generating results based on the retrieved information.

Retrieval-Augmented Generation (RAG) combines search with LLM generation. Instead of asking an LLM to generate answers from its training data alone, we:

Retrieve relevant documents using search
Augment the LLM's context with those documents
Generate responses based on the retrieved information

This gives us (in principle) the best of both worlds:

Accurate information from our search
Natural language responses from LLMs

The RAG Pipeline

The user enters a query.
The search system retrieves relevant documents.
The LLM processes the query and retrieved documents.
- Summarizes information
- Answers questions
The generated response is returned to the user.

Assignment

Implement a RAG pipeline that combines search results with LLM output to produce context-aware answers.

Create a new CLI script at cli/augmented_generation_cli.py. Here's some boilerplate for you to start with:

import argparse

def main() -> None:
     parser = argparse.ArgumentParser(description="Retrieval Augmented Generation CLI")
     subparsers = parser.add_subparsers(dest="command", help="Available commands")

    rag_parser = subparsers.add_parser(
        "rag", help="Perform RAG (search + generate answer)"
    )
    rag_parser.add_argument("query", type=str, help="Search query for RAG")

    args = parser.parse_args()

    match args.command:
        case "rag":
            query = args.query
            # do RAG stuff here
        case _:
            parser.print_help()

if __name__ == "__main__":
    main()

The rag command should:

Load the movies and perform an RRF search using the provided query. Search for the top 5 results.

Prompt the LLM with the query, the search results, and instructions for the LLM to generate an answer based on the retrieved documents. Here's the prompt I used:

prompt = f"""You are a RAG agent for Webflyx, a movie streaming service.
Your task is to provide a natural-language answer to the user's query based on documents retrieved during search.
Provide a comprehensive answer that addresses the user's query.

Query: {query}

Documents:
{docs}

Answer:"""

Print the results and the generated answer in this format:

Search Results:
- We're Back! A Dinosaur's Story
- Jurassic Park
- The Lost World
- Carnosaur
- A Sound of Thunder

RAG Response:
<RESPONSE HERE>

If everything seems to be working, submit the CLI tests.