Spotlights asks the question before the coding agent: which parts of this repo are worth optimizing at all

arXiv·medium signalXBlueskyLinkedInCopy link

Coding agents and evolutionary code-search improve an implementation once you name a target and an evaluation criterion, which leaves the earlier question unanswered on a real repository. Spotlights defines optimization-opportunity discovery, taking a repo, a high-level engineering objective and optional runtime evidence such as telemetry or profiles, and returning candidate source regions with explanations and proposed changes, without the user naming a defect or a bottleneck. It works through logical repository mapping, successive agent reviews and optional research linking candidates to known techniques, evaluated across model serving and document retrieval among other domains.