AI & automation engineer — fintech and reinsurance
I build agent systems that run on their own, and open-source the parts that generalise
A third-line support agent that starts from what a user actually sends you: a screenshot of an error in a web app.
From that image it works backwards to the cause — a bug in the product code, a dependency, or something in the infrastructure — then routes the result by who owns the fix:
| The fix is ours | The fix is someone else's |
|---|---|
| A proposed change, with the failing path identified | The escalation written for them: what broke, what it blocks, what they need to do |
The point is closing the gap between "here's a screenshot, it's broken" and an actionable, correctly-addressed answer — with no human translating in between.
ai-agent-platform — a fully self-hosted, multi-agent LLM platform running an automated coordinator → coder → reviewer → debugger pipeline on commodity, CPU-only hardware. No API bills, no data leaving the box.
Built with Docker Compose, n8n, Open WebUI, LiteLLM and SearXNG.
Tooling for the Agent Skills format, which works with any agent that reads SKILL.md — Claude Code, Codex, Gemini CLI, Cline and others.
Infrastructure
- skills-radar — a curated catalog of agent skills worth installing. 76 entries, every one read by a human before listing, tagged by role. A weekly job scans the ecosystem and ships a digest release.
- role-kits — 23 installable bundles, one command per role. A shared core plus a role layer resolved through kit inheritance, with CI validating every manifest.
- ai-weekly-signal — a Saturday briefing on what actually moved in AI: platforms, models, agent tooling, and the money. Every figure carries a source link.
Skills I wrote
- empathy-in-action — empathy through action rather than words. When emotion is blocking a decision, the assistant picks up the rational load, mirrors the user's register, and never claims a capability it doesn't have.
- read-the-room — reads memes and slang as real instructions instead of noise. Nine speech acts, honest about references it doesn't recognise, and the task still gets done.
Engineer across AI/automation and financial systems, with domain experience in reinsurance — plus hands-on time in project management, QA and business analysis, which is where the interest in support tooling comes from.
Cloud-certified, at home on the command line, and most useful turning a messy manual process into something automated and reproducible.
Languages
AI & automation
Infrastructure
