Intermediate60 minProduction RAG

Document Chunking Strategies

Implement and compare fixed-size, recursive and semantic chunking. Measure the impact on retrieval quality.

Scenario

Your RAG retrieval is poor. Before touching embeddings, experiment with chunking strategies — the #1 cause of bad RAG.

Objective

Implement three chunking strategies and measure their impact on a sample retrieval eval.

Starter code

Implement the TODOs to complete the lab.

chunking.pypython

def fixed_size_chunk(text: str, size: int = 512, overlap: int = 64) -> list[str]:
    """Split text into fixed-size chunks with overlap."""
    # TODO
    pass

def recursive_chunk(text: str, headers: list[str]) -> list[str]:
    """Split by markdown headers, then by size."""
    # TODO
    pass

def evaluate_chunking(chunks: list[str], queries: list[dict]) -> dict:
    """Measure recall@5 and context relevance for the chunking."""
    # TODO
    pass

Solution hints

1Fixed-size: window with overlap
2Recursive: split on headers first, then size within
3Evaluate with the same embeddings and queries across strategies
4Measure both recall and chunk coherence

Validation steps

Your implementation should pass these checks.

Fixed-size produces N chunks with given overlap
Recursive respects markdown structure
Eval returns recall@5 and context_relevance per strategy
Recursive outperforms fixed-size on structured docs

Run validation

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