jaredpalmer/kev

Deploy a Kev decision model (the open Jev-style System One model) as the user's own TypeSafe-compatible HTTPS endpoint on Modal with one command, wire it into their code, and take it down. Use when someone wants to host Kev, get a Kev API URL, replace Jev / TypeSafe calls with a self-hosted model, pick a Kev size or GPU, add an API key, keep an endpoint warm, or stop and remove a Kev deployment.

Extremely strict structural review of a Kev branch or PR (code-judo simplifications, spaghetti growth, files past 1k lines, boundaries, duplication of canonical helpers). Use when asked to review a PR, audit a diff, or run a "thermonuclear" or deep code quality review on this repo.

Search the Kev knowledge graph (private repo jaredpalmer/kev-knowledge) before re-deriving history. Use when asked what was tried, decided, measured or learned in earlier Kev sessions, why a rule or number exists, what a round or PR did, or when starting a research, serving, release or data task that earlier sessions may have covered. Also covers how the graph is refreshed (nightly Devin Cloud automation) and how to add to it.

Launch, monitor and pull Kev training studies, untrained-base probes, remote benchmarks and new-base smoke checks on Modal (modal_app.py). Use when running trials, delta fine-tunes, base probes, external evals or fit checks for the Kev repo.

Verify a Kev change has no regression and ship it as a reviewed PR. Use when refactoring, editing kev/*.py, scripts, the Space or the playground, and when opening or merging Kev pull requests (stacked branches, squash merges).

Write a Kev pull request title and body that teaches the reader why the change exists and how it works. Use when opening, editing or reviewing a PR on this repo, or when the PR title will become the squash-merge commit message.

Fine-tune a Kev decision model (open Jev-style System One model) on a user's own workload and serve it with calibrated probabilities. Interviews the user, finds existing Jev/TypeSafe questions or labelled data in their codebase, generates enough synthetic training data to measure a gain, trains and calibrates on Modal, scores against the released checkpoint, deploys a TypeSafe-compatible endpoint, and tears everything down. Use when someone wants noul/choice/score questions answered on their own domain (routing, triage, moderation, classification, scoring), wants calibrated confidence for automation thresholds, mentions Jev, TypeSafe, System One or Kev, or asks to fine-tune, evaluate, deploy or clean up Kev.

Remove AI-generated code slop and clean up code style

A relentless interview to sharpen a plan or design.

Cut AI tells from any writing. Must always apply.