Applied AI engineer in Vancouver, BC. I build, deploy and evaluate AI agents for real teams.
My own agent is in daily use, and I don't call an agent done until it has an evaluation set, an approval gate for risky actions, and a monitor. The repositories below are the open-source version of that method. All data in them is synthetic.
Portfolio: jaysonqin-dev.github.io
| Repository | What it shows |
|---|---|
| agent-playbook | An evaluation harness that diffs two runs to show what a change fixed and what it broke, a weekly monitor, and the lessons behind them. |
| agent-tool-safety | Safe writes for tool-using agents: risk tiers, dry-run, approval bound to the exact plan, tamper-evident audit log, undo. |
| extraction-agent-demo | A parity gate for swapping the model behind a document-extraction step, with a cost model for the decision. |
| bookmark-organizer | Marksly: a Chrome extension that finds misfiled bookmarks with an 8 MB on-device model. Live demo. |
- Eval lab: read the diff between two agent runs, and find the case that quietly got worse.
- Learning loop: judge which corrections should become rules for an agent.