I build AI agents that ship software and audit money.
Short engineering stories, not pitch decks. Open a card for the full write-up; the URL lands right on it.
A personal-finance product I run in production and build alone. One schema across a React web app and native iOS and Android clients, bank statements pulled from Gmail and parsed by Claude with the user's own key, a data pipeline with a quality dashboard, and an AI team that ships changes as tested pull requests.
Why deterministic checks kept missing the long tail, how a 6-agent LangGraph pipeline closed it, and the messy parts: hallucination guardrails, retry loops, a self-learning rule memory, and a golden-set harness that keeps the LLM honest. The same pattern now audits my own finance data in ZevMon.
CloudFront, S3, Lambda and DynamoDB, all in Terraform. A chat grounded in my work with voice in and out, and a Telegram-backed contact form.
Query latency from 30s to 8s, enrichment quality to 99.97%, and seven teams trusting the same gold tables.
Month-end spreadsheets became next-day automated reports; detection went from 30% to 99.99%.
Longer writing on what I'm building and learning.
I'm a data and AI engineer based in Waterloo, Ontario. For nine years I've built big-data platforms in banking, insurance and automotive, from 200B+ event pipelines to the semantic layers on top, and I currently lead work on LLM multi-agent systems that analyze $1.5B in transactions.
I've also taught since 2009, as a university lecturer in networking, security and algorithms, and I still mentor engineers. Away from the keyboard: poetry, science, music and philosophy. The chat in the corner will happily talk about those too.
Hiring, collaborating, or just curious? Email nxv.can@gmail.com or use the chat.
AI-generated answers grounded in Vinh's resume & this site. May contain errors. 🔊 Listen uses Vinh's cloned voice (ElevenLabs) with auto EN/VI detection.