Projects

Three platforms, one pipeline shape

Each one exists so I could practice a specific engineering problem at production standard. The domains are deliberately kept out of focus. The engineering is the point. Pick a project below.

Personal, non-commercial learning projects. Built solo, outside work hours, and not operated for any client or company. All numbers below are from the repositories as of October 2026.

Agentic SDLC Pipeline

claude code · gemini · copilot · github actions The one that built the others

A custom multi-agent workflow that takes a requirement through clarification, specification, design, implementation, tests, review and documentation, with a human gate between phases. Every project below was built through it. The honest lesson after ten months: the agents are only as good as the gates around them, and most of my effort went into the gates.

How it works

  • Each repo carries the agent contract: a working-agreement file with hard constraints, a Definition of Done, and phase templates for analyse, implement, test, review, document and close.
  • Agents must ask for clarification before drafting a design. No spec, no code.
  • Commit messages, spec traceability trailers and ticket references are checked by hooks, not by trust.

What keeps it honest

  • Coverage ratchets that can only go up.
  • Mutation-testing floors per module.
  • Architecture tests and about thirty policy scripts that fail the build on things like floating-point money or an uninjected clock.
  • Docs regenerated by CI and proposed as a pull request, so documentation drift is a diff, not a surprise.
Nothing deploys unless CI is green for that exact commit. Staging first, production by hand.
Nothing deploys unless CI is green for that exact commit. Staging first, production by hand.
10 monthsin daily use since Dec 2025
4,000+commits across three repos
4 languagesJava, TypeScript, Python, Dart
3nightly gate suites: mutation, load, 90-day clock