I am a Software Engineer on the Upstage Data Team.
My immediate focus is exploring how we can apply AI agents to solve production-level problems.
Here’s a summary of my recent work:
- Building systems that help people better evaluate and guide LLMs. This includes automated qualitative evaluation frameworks, LLM-assisted human feedback collection platforms that reduce data creation difficulty while improving data quality, and agentic tools that bridge the gap between traditional metrics and real-world performance.
I also worked on applying RAG to build a chat backend for virtual K-pop idol MAVE:, serving thousands of users worldwide, as well as an automated insurance evaluation pipeline for Hanwha Life.
Before joining Upstage, I was an AI Researcher at Silvia Health, where I trained models to detect cognitive impairments from speech on smartphones.
I completed my Master’s in the Interactive Computing Lab at KAIST. I studied active learning to leverage scarce emotion data collected in everyday settings for emotion recognition (see AdaptiveESM), and created the K-EmoCon Dataset, a multimodal biometric sensor dataset for recognizing emotions in the wild.
I’ve also been dabbling in Web3, occasionally trading and building.
Broadly speaking, I’m interested in this question:
How can we correctly incentivize data creators so we can have more “good” data?
Earlier this year, I built alignment-protocol, an incentive alignment system for AI data for a Solana hackathon. You can interact with the protocol at ArVxFdoxzCsMDb1K3jXsQTrDP4mbfHMxKiZLjZpznB5c on Solana devnet.
Recently, I’m spending most of my free time vibe-coding various apps for myself.
I’m usually at climbing gyms when I’m not in front of my laptop.