8000 jeonghun917 (Jeonghun Mun) · GitHub
[go: up one dir, main page]

Skip to content
View jeonghun917's full-sized avatar

Block or report jeonghun917

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
jeonghun917/README.md

Jeonghun Mun

I built a control plane in the first place because of shitty agent memory and constant session changes. 140 PRs later, the part that worked was the simplest one, so I shipped it on its own:

readfirst: what to read and check before you change a file

One file next to the code holds the rules, traps and checks only your team knows. npx readfirst read <file> hands them to you or your coding agent right before an edit (a Claude Code hook, MCP for Cursor and Codex, or Cursor rules exported from the same map), and a GitHub Action fails the PR when the map falls behind the code. readfirst find shows what already exists before an agent writes it again; readfirst audit shows what the map no longer sees. MIT, no dependencies, no model calls. Making maps for open source projects turned up real bugs, reported to saleor, wagtail, Tandoor and mealie. Site · npm · Marketplace

Welsmere: a British-English speaking coach for Korean and Japanese learners

Learners talk by voice: role-plays of everyday and work situations, goal-based lessons, a daily UK course for living in Britain, IELTS mock tests, and spaced review of their own sentences. Every talk ends in a result sheet with corrected sentences, expressions to try, phoneme-level pronunciation feedback with their intonation drawn against the coach's British voice, and a rough CEFR or IELTS level. Live, with paying users.

TTS Blind Eval Studio: blind listening tests for text-to-speech

A small CPU-only tool for comparing TTS candidates without knowing which model made which clip. Candidate names stay hidden until the reveal, labels and order are randomised from a seed, and the prompts and rating rubric are locked to a versioned protocol so they can't drift between evaluations. No model weights, datasets, API keys or GPU needed.

Current work includes:

  • AI Development Control Plane — continuity, bounded execution, validation, review, and promotion
  • Ars Mentis / VELA — persistent AI continuity
  • Pharos Orbis — provenance-grounded information systems
  • C3 — speech and interaction systems
  • Ars Bellum — simulation and game systems

Public notes:

Most production repositories are private.

Pinned Loading

  1. readfirst readfirst Public

    What to read and check before you (or your AI agent) change a file: the project's rules, risks and tests, tied to the code.

    JavaScript 5 1

  2. ai-development-control-plane ai-development-control-plane Public

    How I ended up building a control plane for long-running AI-assisted software development.

    1

  3. tts-blind-eval-studio tts-blind-eval-studio Public

    Python 1

0