Team AI Engineering Practice

A lot of senior-engineering effort was going into work that repeated: the same kinds of review feedback, the same manual gathering of evidence from scattered sources. It was a tooling gap, not a people problem.

1. 01

   Authored a Claude Code slash-command suite that's now in production use by the team — codifying recurring workflows into commands anyone can run.

2. 02

   Built an MCP-driven, multi-source evidence pipeline so context that used to be assembled by hand is pulled together automatically.

3. 03

   Ran a quantitative analysis of a 1,000-merge-request review corpus to find what reviewers actually repeat, then proposed pre-merge tooling that automates roughly one-sixth of recurring review feedback.

4. 04

   Drove a session-replay rollout with privacy designed in from the start: a whitelist-based opt-in, automatic-deployment steps deliberately removed to prevent unintended production recording, and a competitor-disclosure analysis that shaped the privacy-policy language.

AI tooling I authored is in daily team use, and the review-automation proposal is grounded in a measured corpus — a defensible estimate of impact rather than a guess.