Senior Generative AI R&D Engineer · Audio & Music AI Developer · Independent Musician
Building practical AI systems and audio software for creators, communities, and intelligent workflows.
I build practical AI and audio software that turns research into working tools for creators. My recent work combines local ONNX inference, C++ and Flutter desktop applications, DAW integration, source separation, melody transcription, live pitch tracking, and cross-platform audio plugin development for AAX, VST3, and AU.
Alongside music technology, I develop production-facing generative AI systems—including RAG and agent workflows, structured outputs, Physics AI, and multilingual knowledge products—with an emphasis on traceable automation, cross-platform delivery, and interfaces people can use without machine-learning expertise.
Open to collaborations in generative AI, audio software, creator tools, and music technology.
Generative AI RAG, agents, structured outputs, traceable workflows
Music AI Source separation, transcription, pitch and tempo analysis
DAW tooling Pro Tools PTSL, stem workflows, AAX, VST3, AU
Desktop products Flutter, C++, ONNX Runtime, macOS and Windows packaging
Web systems Astro, React, TypeScript, FastAPI, GitHub Pages
| Project | What it does |
|---|---|
| TrackMate |  An AI-assisted Pro Tools stem-import workflow that recognizes instruments from filenames and audio, standardizes track names, categories, and colors, detects BPM and tempo maps, identifies fake stereo for mono conversion, and imports and verifies everything through Avid's official PTSL—all locally. |
| karaoke-battle | A local-first karaoke scoring app for streamers that turns YouTube links or media files into singable tracks with vocal separation and melody transcription, displays live pitch against scrolling notes, and scores pitch, rhythm, stability, vibrato, dynamics, and sustained notes on macOS and Windows. |
| ai-interview |  A multilingual AI engineering interview platform with source-linked flashcards in Traditional Chinese, English, and Japanese, structured AI-assisted study notes, community contributions, and daily incremental updates via GitHub Actions. |
| PIE — Physics Informs Everything (private) | A PhysicsNeMo-powered MCP bridge and nine-skill research workflow that helps AI agents formalize physics problems, inspect models and literature, process CAD and mesh inputs, validate PhysicsNeMo implementations, and run traceable experiments. |
| music-detection | A privacy-first browser music analysis tool that detects BPM, time signature, key, and vocal pitch range with ONNX inference, previews beat and pitch results, and exports Tempo + Lead Vocal MIDI, with an optional local Python engine. |
| HT-76 |    A JUCE-based 1176-style FET compressor for macOS and Windows, with Rev D and experimental Rev H circuit models, VST3, AU, and AAX builds, plus switchable classic rack and real-time analysis views. |
| JS_Inflator (customized fork) | A cross-platform JS Inflator fork with macOS and Windows VST3, AUv2, and AAX builds, installer packaging, and audio/UI regression testing; its AAX adaptation targets Pro Tools Developer and remains in testing. |
| software-salary | Taiwan software salary insights based on 709 analyzed salary samples, with market percentiles, company rankings, experience-based estimates, and interview guidance. |
| Private RAG / Agent Systems (private) | Internal AI systems for retrieval-augmented generation, document workflows, structured outputs, and production-facing automation. |
| dc-manager | An LLM-powered Discord community assistant for rule-violation detection, configurable moderation responses, channel analytics, activity summaries, member tracking, and Traditional Chinese operations. |
- 2026-11-07 — Updated Hikari-Tsai/ktv-charts: Updated music metadata and karaoke charts with new track matches, audio features, and refreshed daily rankings.
- 2026-11-07 — Updated Hikari-Tsai/ai-interview: Refreshed several interview QA entries with updated sources, clearer wording, and newly ready localized answers.
- 2026-11-07 — Updated Hikari-Tsai/ai-werewolf: Added configurable night kills and delayed reveal, plus UI/docs updates for last words and GitHub log sharing.
- 2026-11-07 — Updated Hikari-Tsai/karaoke-battle: Raised scoring thresholds and documentation to use the new lenient scale, and refreshed README screenshots and lyrics UI text.
- 2026-11-06 — Updated Hikari-Tsai/karaoke-battle: Updated the readme to better explain singing visuals, scoring, and lyrics import/editing features.
- 2026-11-06 — Updated Hikari-Tsai/ai-werewolf: Added contribution docs and polished the lobby UI to show the room code, refresh static assets, and better prompt voting reasons.
Last updated: 2026-11-08 09:31 (UTC+8)
Public repositories 35
Admin private org repos 6
Total stars 134
Last-year activity 648 contributions
Merged pull requests 65 public PRs
Current achievements Galaxy Brain, Starstruck, Pair Extraordinaire, Public Sponsor, YOLO, Pull Shark x2, Quickdraw
Highlight GitHub Developer Program Member
Last updated: 2026-11-08 09:31 (UTC+8)
 Guardian |
 500 Days Badge |
 Dynamic Programming II |
 Algorithm II |
 Data Structure II |
 Programming Skills II |
 LeetCode 75 |
 Graph Theory I |
Show 46 more badges
|
|||
| Area | Technologies |
|---|---|
| Core Engineering |
|
| AI & ML |
|
| Web |
|
| Audio & Music AI |
|
| Plug-in Development |
|
| Automation & Platforms |
|
Building practical AI and audio software for creators.