Projects

Turns regulations into concrete actions for companies, using AI agents.

Decision

Designed and specified with arc42: quality goals first, domain core isolated from the AI agents via ports, so models can change without touching business logic.

Built

Architecture, backend, AI agents, frontend and cloud infrastructure.

Lets developers extend a multi-repo platform from day one, guided by their AI agent.

Decision

Architecture rules and patterns shipped as agent skills instead of docs, so every change follows the same design, whoever writes it.

Decision

Tool-agnostic: works with Claude Code, OpenCode and Pi alike.

Built

Skills, patterns and the harness that delivers them.

One embedded codebase, several ECU targets, several teams.

Decision

Cut the codebase along layer boundaries into Gradle modules, each with one clear owner.

Result

Faster builds, lower cognitive load for the teams, independent delivery per target.

From no CI at all to one delivery pipeline shared by all teams.

Decision

A shared library instead of per-team scripts: preflight, build, test, encrypt, analyse and deploy as reusable building blocks.

Result

Delivery scaled with the teams; took over ownership of the entire CI/CD.

Tools that bridge systems in CI and let my teams move faster.

Decision

Clean Architecture as a fixed frame: tools stay maintainable and extendable, and juniors grow into senior thinking by working within it.

Decision

Generated SDKs instead of hand-written clients: teams build against an evolving backend without waiting.

Result

Over 90% test coverage, OAuth-ready clients in a few lines of code.

One channel for all project communication in an Agile Release Train.

Decision

Communication as code: every message is a merge request, reviewed, versioned and rendered into a consistent email design.

Result

Replaced scattered Teams, email and Confluence posts; ART leads can steer and intervene before anything goes out.