auto-review-loop-llm

Automate iterative review by configuring an autonomous LLM loop that evaluates ML research artifacts, applies fixes, and re-checks until reaching a positive assessment or MAX_ROUNDS limit. It guides manuscript and evaluation plan improvements using structured feedback.

An autonomous LLM review loop iteratively evaluates ML research artifacts by applying structured feedback, implementing fixes, and re-running reviews. It converges toward a submission-ready result while persisting state locally in REVIEW_STATE.json and AUTO_REVIEW.md.

Yes, you need a configured llm-chat MCP server to run automated ML artifact reviews. You must also specify your provider and model in ~/.claude/settings.json to enable the LLM-guided multi-round evaluation process.

Configure MAX_ROUNDS and POSITIVE_THRESHOLD to control the iterative feedback loop's depth and convergence criteria. These parameters determine how many review-fix cycles execute before stopping and what assessment score qualifies as a positive result.

Yes, you can use automated review loops for both academic and industrial ML projects. The iterative evaluation, fixes, and re-checks apply to research papers, reports, and code artifacts to converge toward submission-ready results.