Profile
I'm a Computer Science student from UC Berkeley and National Tsing Hua University, working at the intersection of machine learning for music and audio, LLM alignment, and HCI.
My research lives between BAIR under Prof. Trevor Darrell and David M. Chan, and the CNMAT under Prof. Carmine-Emanuele Cella.
My long-term goal is to build audio-language models that can truly hear and understand music. This exposes a brand-new human computer interaction method to musicians, to discuss, collaborate, or even interact with non-musical audiences.
Before research, I was a child musician: 24 solo recitals and more than 70 TV appearances across China and Taiwan between 2009 and 2016. I still play, write, and produce. That dual lens of engineer and musician is the through-line of everything I build.
Publications & Research
Music AI has never looked more promising nowadays: companies like Suno, Udio, and Ace Studio show how generative models may reshape music itself. However, I'm deeply concerned that much of this field is moving toward replacing the humanity, intention, and creativity from the soul of a musician.
I tried to reflect what was missing and how to leverage the advantage of this exploding generative AI era. My research starts from three missing pieces:
Many models nowadays can produce impressive zero-shot results, but giving very little control to musicians.
Many models learn patterns at latent space level from large-scale internet data, but still lack of grounded understanding
from “perceptual level” (pitch, rhythm and music structure),
to “hidden level” (the intention from the composer, the expression from the player, and the emotion from the audience).
But these are how we bring music from random permutation of notes to life.
Music is the most abstract language we have. There is no ground truth hidden in the notes or audio signal itself. The same musical phrase can bring huge variance between individuals. This leads me to a research question I care deeply:
can ALMs model and align human preferences?
Together, these questions shape my long-term research direction: building a unified audio-language model that can truly hear, understand, and communicate about music, enabling new forms of HCI between musicians, computers, and audiences.
Research · Industry · Leadership
BAIR, UC Berkeley · advised by Prof. Trevor Darrell & David M. Chan
CNMAT, UC Berkeley · advised by Prof. Carmine-Emanuele Cella
National Tsing Hua University · advised by Prof. Yi-Wen Liu
Guitar amp modeling & audio ML in production.
ASUC's flagship student platform · 122K+ monthly users at UC Berkeley.
Pianist · Composer · Producer
I learned piano by ear. My first competition (a national piano contest in Taiwan), I won first place without a teacher. What followed, between 2009 and 2016, was a child-star decade: more than 70 TV appearances, 24 solo piano recitals across mainland China and Taiwan, three on-stage moments with Jay Chou, and a featured guest spot at Yen-J's Taipei Arena concert.
I never stopped writing. I have 20+ original songs, and my single “失電陣線聯盟” crossed 14.4K plays on Instagram. In college I played keys, lead vocal, cajón, and electric guitar across several bands; whatever the song needed, I'd fill in. If it's keyboard, I don't need rehearsal: one listen and I can improvise the part on stage.
Go Bears
Outside research, I designed and built two presentations to capture this year: one for our Berkeleytime ML Pod's final showcase, one as a personal scrapbook of my exchange.
ML Pod · OCTO Showcase
Designed and built the Berkeleytime ML Pod's end-of-semester showcase deck: Explore page launch, semantic search infrastructure, and pod leadership over Spring 2026.
Open presentation →
Personal Scrapbook
A handcrafted slide deck of my Berkeley exchange year: BAIR, classes, frat, trips, and the people who made this the best year of my life.
Open scrapbook →Awards & Hackathons
Code & Side Projects
A subset of work from github.com/vaclisinc. Roughly grouped: audio & ML, hackathon products, open source, full-stack, and hardware. The full list lives on GitHub.
JUCE · C++ · PyTorch · ReaScript
A JUCE-based VST plugin integrating my audio-to-preset (VTR) model for real-time EQ parameter prediction inside any DAW. Companion to the arXiv paper.
github.com/vaclisinc/VTR-smartEQ →Python · PyTorch · DSP
The underlying audio-to-preset model: predicts multi-band EQ parameters directly from a reference recording. Trained on a ReaScript-automated piano dataset.
arXiv 2509.24404 →Python · CLAP · LLM
Interactive demo for multi-turn, text-guided audio effect refinement. Parses feedback like “less bright / more warm” into directional updates in CLAP embedding space.
github.com/vaclisinc/InstructFX2FX →Python · PyTorch · RNN
An RNN-based pitch-type prediction system analyzing MLB pitching sequences to uncover patterns and refine batting strategy. Bridge between baseball analytics and sequence modeling.
github.com/vaclisinc/Pitch-Type-Prediction-Using-RNN →Vue 3 · Flutter · Taipei Parking API
Real-time parking, navigation, and notifications for Taipei. Selected for integration into the official TownPass super-app (3M+ downloads). NT$100,000 prize.
github.com/vaclisinc/ParkFlow →Flutter · Python · Gemini · Google Maps
AI personalized travel planner built around interests, personality, and real-time data. 2nd Place (Google) at Meichu Hackathon 2024.
github.com/vaclisinc/Travelity →Flutter · Python · OpenAI · LangChain
An AI-powered mental-health diary with anonymous social features. 1st Place at SITCON Hackathon 2024: built with the team to make journaling less lonely.
github.com/SimonLiu423/cc_diary →TypeScript · GraphQL · MongoDB · Redis
ASUC's flagship student platform serving 122K+ monthly users at UC Berkeley. I lead the ML pod and have shipped 20+ PRs to the production codebase as a full-stack engineer.
my PRs →JavaScript · IMAP / SMTP
Contributed an execution-trace notification system and terminal-style email UI; fixed a critical self-reply loop. The contributions trended on X with 126K+ views.
github.com/vaclisinc/Claude-Code-Remote →React · Python · FastAPI · Firebase
A web platform of subservices for amateur baseball: including jyBaseball, a community-driven digitization tool. Built with the NTHU baseball team and Taiwanese baseball YouTubers.
github.com/vaclisinc/InningIQ →Verilog · FPGA
An FPGA dual-mode cleaning robot combining manual flexibility with autonomous cleaning. A full hardware/software co-design exercise on an Altera board.
github.com/vaclisinc/PilotClean →Python · Raspberry Pi · AWS
IoT-based interactive recycling reward system using a Raspberry Pi camera and AWS cloud services. Sustainability hack with real hardware deployed in a campus pilot.
github.com/vaclisinc/RecyScore →Get in touch
Actively seeking 2027 Fall PhD positions.
I'm happy to talk about audio-language models, music technology, hackathon ideas, piano performance, or anything in between. The fastest way to reach me is email.
vaclis@berkeley.edu