Tey Y.

I build the parts of AI agents that make them reliable over time: what the agent remembers, how its answers are judged, and how you prove it still works after every change. At Grab, as a Senior AI Engineer, I work on evaluation for customer-support AI agents: - An annotation UI that lets domain experts review conversations and agent flows - A data pipeline processing 200k+ support conversations a day - Step-level latency tracing across multiple agents, LLM and API calls - LLM-as-a-judge audits of production conversations for SOP compliance I have also built agent memory systems end to end: - Extracting memories from conversations, merging duplicates, and resolving conflicts when a user corrects themselves - Deciding what is worth remembering, and validating evidence before anything is written - Evaluation and replay harnesses that test memory behaviour on real cases before release - MCP connectors and Claude Skills so agents can read and use that memory What I can help you with: - Your agent gives wrong answers or ignores its tools, and you need to find out why - Designing long-term, per-user memory or context for an AI product - Building evals, LLM-as-a-judge checks and regression tests for LLM features - Claude Skills, Claude Code and MCP integrations, including permissions and human approval - Architecture reviews and coaching for teams already building I work part-time, with clear written updates. English and Chinese (native).