Learned file ranker without an LLM

The idea

Gradient-boosted trees rank a request's candidate files from about 140 counted signals and the Git history, with no LLM call.

The problem
Text search missed many of the files a change needs.
Why it matters
On held-out change requests it found 39–54% more of the right files in the top 10 than the text-search setup.
Where it stands
Shown in a lab run as of

A step-by-step replay of this idea is not published yet. The library adds one when a recorded run shows it.

Where it lives in the product

  • Code knowledge indexes for agents

    A code index at the level of signatures finds code in milliseconds. An Atlas profile, computed by code with no model, gives each repository's stack, commands, rules and module map.

Turn the numbers into shipped work.

Crewshal runs these choices for you: a persistent AI worker with memory and rules, on your Claude and Codex subscriptions.