BM25

What it means

It scores documents based on term frequency and length. In retrieval workflows, bm25 often shapes grounded answers and search relevance.

How it works

Retrieval pipelines index content into chunks and embeddings, then fetch relevant pieces at query time. The model uses those snippets as context to answer.

Why it matters

Retrieval improves accuracy by grounding responses in your data.

Common use cases

• Question answering over internal docs or knowledge bases.
• Support assistants that cite sources and reduce hallucinations.
• Enterprise search that understands intent beyond keywords.

Example

Rank docs for token budget.

Pitfalls and tips

Poor chunking or stale data leads to irrelevant results. Refresh indexes and tune chunk size to keep answers accurate.

In BoltAI

In BoltAI, this appears in search, knowledge, and grounding workflows.