Yu Han Tey
AI agent engineer · Singapore
I build AI agents that remember things about their users, and the tests that show whether they get it right.
Most of my work is on:
- Agent memory: built a memory system end to end. It decides during a chat what is worth keeping, runs a nightly job that merges and updates memories, lets users correct them, and imports memories from other AI assistants. Most steps have their own eval, so a change in one step can't quietly break another.
- LLM evaluation and observability: tools to replay a failed run step by step and label what went wrong, so debugging takes minutes instead of hours. Model judges that score conversations against a checklist, so quality is measured, not guessed.
- Agent skill design and evals: laying out skills so an agent opens the right ones, then testing two things separately: did it pick the right skill, and did the answer actually get better.