I make multi-agent and agentic coding systems safe by design, turning abstract policy goals into deterministic constraints, not prompting. I contribute correctness/trust fixes to the open-source agent tools, and I'm working on an eval harness that judges agentic-coding traces (diff quality, tool choice, task completion) to make agent competence more quantifiable. I'm judicious about what gets automated and what stays human-in-the-loop.
🟢 Open to remote contract work · agentic-coding reliability: correctness fixes, trace evaluation, and guardrails for teams shipping coding agents. ~40 hours/week, async-first, UTC+7 with EU/US-morning overlap. Email: minh2416294@gmail.com
-
Build Log: https://minh2416294.github.io/projects/
-
Open-Source Contributions: https://minh2416294.github.io/contributions/
-
Engineering Notes: https://minh2416294.github.io/posts/