Projects

Things I've built and shipped, from Gemini audio launches to multilingual LLMs and audio-visual speech models.

Google DeepMind

On the Gemini Audio team I work on speech and audio understanding for Gemini, from core model capabilities to safety alignment for audio-in and audio-out launches. Launches I contributed to:

Earlier
Trillion Labs2024 – 2025

Tri-21B & Trillion-7B

As a founding team member, worked on model architecture and training infrastructure for multilingual LLMs built for real-world deployment, part of Trillion Labs’ effort to build Korea’s most powerful foundation model. Tri-21B was released open source.

LLMpretrainingmultilingual