Junha Park

Hi, I’m Junha.

I work on evals for LLM agents. I'm especially interested in fully autonomous coding agents for OSS maintenance and long-horizon software engineering. Recently, I authored a Terminal-Bench 3 task and am building a new benchmark to evaluate coding agents' ability to make long-term architectural decisions.

Before that, I built FastRoll, a validator node for the JAM protocol, and passed the JAM Prize Milestone 1 conformance fuzzing. JAM is a decentralized computing protocol designed to replace Polkadot's relay chain, which secures over $1 billion in digital assets. I also contribute to the Gray Paper, JAM's formal specification, as a protocol reviewer.