Khang Nguyen Khang Nguyen

AI workflow portfolio

Hi, I’m Khang!

I build practical solutions with AI agents and human review.

View work Contact

Work that can be checked

No inflated metrics. No pretend production claims. Each item needs a repo, screenshot, command output, or live URL.

Live now

AI workflow portfolio

This site documents how I use agents, where proof exists, and where claims still need live evidence.

View project

Agent harness engineering mechanism

I use the harness idea in daily work: humans steer, agents execute, and the system checks outputs before anything ships.

  1. Intent

    I define the user goal, constraints, source links, and acceptance checks before agents touch the task.

  2. Workbench

    Agents operate inside a bounded tool setup: repo files, browser checks, scripts, search, screenshots, and deploy commands.

  3. Gates

    Each run must produce proof: diff, command output, visual check, source trace, or live URL. Failed checks loop back into fixes.

  4. Review

    I inspect the result, cut unsafe claims, decide what is publishable, and keep only work that can be verified.

Let’s build something useful

If you need a practical solution, workflow prototype, or AI-assisted build with real verification, send me the context and goal.