Experience

Research positions, engineering work, awards, invited talks, music tools, and open-source contributions. Papers are listed on the home page.

Research

  1. NTU Music and AI Lab

    • Leading the development of a large audio-language model (LALM) for music, bringing together insights from prior music benchmarking, methodology, and diagnostic studies into large-scale model alignment and training, with collaborators from NVIDIA, BAIR, NTNU, Queen Mary University of London, and Prof. Hung-yi Lee's group.
    • Researching symbolic representation alignment for text-to-music generation and representation transfer and utilization in audio-language models, leading to MIDI-REPA and encoder–decoder probing studies.
  2. Berkeley AI Research

    • Revisiting the ubiquitous 25 ms log-Mel frontend used across general audio encoders and models under conflicting speech–music resolution demands, leading to UniFront.
    • Lead-developed PitchBench and Anamnesis, spanning the basics of music understanding in audio-language models and large-scale human simulation.
  3. Center for New Music & Audio Technologies (CNMAT)

    • Led the research direction on interactive, musician-centered audio effect modeling and editing, extending single-shot methods toward iterative refinement and leading to InstructFX2FX.
  4. Acoustics and Hearing Group (AHG)

    • Developed the audio-to-equalizer model (VTR) for single-shot tone replication as my undergraduate capstone project, later deployed as a Virtual Studio Technology (VST) plugin at Positive Grid.

Selected Engineering Experience

  1. BerkeleyTime

    Machine Learning Pod — Pod Lead Jan – May 2026

    • Led a 10-member machine learning team across semantic search, recommendation system, and enrollment prediction, coordinating researcher–engineer pairs and designers from model development through deployment.
    • Led development of a new Explore page, integrating course recommendations into BerkeleyTime's product experience.

    Catalog Pod — Machine Learning Engineer Sep – Dec 2025

    • Built BGE text-embedding and FAISS vector-search services with FastAPI for 122K+ monthly users.
    • Reduced GraphQL first-load latency from 25s to 1s with pagination and Redis caching; shipped and fixed 20+ pull requests.

    PresentationPull requests

  2. Positive Grid

    • Served as Scrum Master for Amp-AI-SaaS and designed the Multi-Agent Orchestration Harness for ongoing software-engineering workflows at Positive Grid.
    • Integrated my capstone VTR audio-to-EQ model into a JUCE VST plugin, using PyTorch inference for real-time audio processing in a digital audio workstation.

    Multi-Agent Orchestration Harness (PDF)

Awards & Invited Talks

  1. Meichu Hackathon 2026

    • Built GodEyes, a physical AI system using a world model to reconstruct crime scenes as explorable 3D worlds, with SLAM-based navigation and on-device NPU gesture tracking on an NXP i.MX 93 edge board.

    World model (World Labs Marble) · 3D Gaussian Splatting · SLAM · Edge NPU · NXP i.MX 93

    GodEyes workflow: crime-scene evidence, AI reconstruction, and viewpoint verification
  2. Claude × a16z × Berkeley M.E.T. Makeathon

    • Built Haven in five hours, a mobile app that personalizes home-design recommendations from room scans.
    Haven home-design app interface
  3. Meichu Hackathon 2024

    • Built Travelity, an AI travel planner using interests, personality, and real-time map and search data.

    Flutter · Python · Google Maps API · Gemini

    Travelity team receiving second place in the Google group at Meichu Hackathon 2024
  4. Taipei Hackathon 2024

    Vue 3 · Flutter · Taipei Parking API · Mapbox

    ParkFlow and fellow winners at Taipei Hackathon 2024
  5. SITCON Hackathon 2024

    • Built ccDiary, an AI-assisted mental-health journal with emotion analysis and anonymous social features.

    Flutter · Python · LangChain · PyTorch

    ccDiary team with their SITCON Hackathon 2024 award
  6. WorldQuant Challenge 2024

    WorldQuant International Quant Championship 2024 certificate for Song-Ze Yu
  7. NCKU Campus Speaker

    • Delivered an invited talk for incoming students on writing music as a diary, then turned audience-submitted phrases and sentences on Slido into improvised songs performed live.

Music Tools GitHub ↗

  1. npm package

    MIDI RealPlayer — Web

    • Designed for music researchers building demo pages and listening tests. Embed synchronized audio and MIDI playback with per-track controls directly in your page.

    Live demonpmGitHub

  2. VS Code extension

    MIDI RealPlayer — VS Code

    • Play and inspect multi-track MIDI inside VS Code, with SoundFont instruments, synchronized piano-roll visualization, and track mixing controls.

    VS Code MarketplaceGitHub

  3. Music tool

Open Source

  1. Education

    BerkeleyTime

    • Led a 10-member machine learning team across semantic search, recommendation system, and enrollment prediction, coordinating researcher–engineer pairs and designers from model development through deployment.
    • Led development of a new Explore page, integrating course recommendations into BerkeleyTime's product experience.
    • Built BGE text-embedding and FAISS vector-search services with FastAPI for 122K+ monthly users.
    • Reduced GraphQL first-load latency from 25s to 1s with pagination and Redis caching; shipped and fixed 20+ pull requests.

    GitHubPull requests

  2. Civic tech

    ParkFlow — Taipei TownPass

    GitHub

  3. Agent tooling

    Claude Code Remote

    • Contributed improvements to reliability and configurability for remote Claude Code workflows, with contributions highlighted on X.

    GitHubPull requestsPost