Skip to content
View mchen04's full-sized avatar

Block or report mchen04

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. Markdown is supported. 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
mchen04/README.md

Michael Chen

AI software engineer at Epoch/ZergAI. CS graduate, UC Riverside (June 2026), with a minor in Entrepreneurship & Strategy.

Open source

  • mlx-audio — fixed five divergences between the MLX iSTFTNet decoders (Kokoro, KittenTTS) and the PyTorch reference they were ported from, found by diffing every intermediate tensor across both stacks on identical inputs. Kokoro rendered 2.5 dB quiet (istft overlap-add normalized by Σw instead of Σw²), and the pitch/energy upsample path was misaligned one frame against its own shortcut (F0 relative RMSE 0.134 → 0.0000). Validated over 55 utterances against frozen fp32 reference audio with thresholds calibrated to the reference model's own render-to-render noise floor: MCD 7.29 → 4.09 dB, F0 RMSE 11.5 → 5.1 Hz. Merged upstream, with regression tests.

Background

  • Epoch — AI SWE Engineer: coding agents, terminal tooling, evaluation systems, multi-agent simulations.
  • MOBIVOLT LLC — Software Engineer Intern: fuel-cell test automation, hardware simulation, firmware migration support.
  • AI at UCR — Founder and President of UCR's first official AI organization (50+ members, 10+ workshops).
  • UC Riverside — B.S. Computer Science, Dean's Honor List.

Links

Portfolio · LinkedIn · HuggingFace

Pinned Loading

  1. PHiLIP PHiLIP Public

    AI image generation and enhancement pipeline using generative models with configurable style prompts and automated generation loops.

    Python

  2. kestrel-tts kestrel-tts Public

    Kestrel: distilled frame-rate TTS for Apple Silicon — Kokoro-82M made 57x faster on an M2 (MLX), with an honest quality report

    Python

  3. claude-switch claude-switch Public

    macOS CLI for sub-second switching between Claude Code accounts via Keychain swap, with optional shared config and per-profile usage dashboard.

    Rust

  4. tcd tcd Public

    zsh + tmux helper to launch, find, and reattach to long-running AI CLI sessions (Claude Code, Codex) by partial name — survives reboots and phone reconnects.

    Shell

  5. hark-audiobook hark-audiobook Public

    Private, installable offline-first audiobook PWA with on-device MP3 storage, chapter-aware playback, cross-device progress sync, and no audio uploads.

    TypeScript

  6. epub-listener epub-listener Public

    Python CLI that converts EPUB books into chaptered MP3 audiobooks using Edge-TTS, Kokoro, Hugging Face models, or local narration commands.

    Python