I start from the business problem, not the technology.

I've spent 12+ years shipping clean, well-architected systems end-to-end, and 8 years building AI/ML since 2018 — from NLP, speech, and recommendation models to, over the last three years, generative AI: multi-agent orchestration, RAG, fine-tuning, and real-time voice.

The thing I actually optimize for isn't the fanciest stack — it's the solution that genuinely fits. Sometimes that's a multi-agent system; sometimes it's the simplest, semi-automated thing that quietly solves the problem. I stay close to the customer and the business logic, and I learn fastest by shipping real solutions for real users. This log is where that shows.

What I'm building now

Two production GenAI systems, both built end-to-end: an autonomous multi-agent engineering platform where AI agents design, build, and deploy web apps; and a multi-channel AI agent ecosystem running live for a 15-venue F&B group (~700 staff). I also built an AI-receptionist voice demo solo.

The shape of the work

Also: I break things on purpose

I've been a responsible-disclosure security researcher since 2019 — I flagged a large user-data exposure at a major Vietnamese social network, and I run bug bounties on HackerOne across crypto-wallet, mapping, and CMS programs. Thinking like an attacker makes me build sturdier things.