Senior platform engineer and technical lead working on infrastructure automation, developer tooling, and AI-assisted engineering systems.
I stay close to the code. Most of my work is around Kubernetes, Kafka, cloud infrastructure, internal developer platforms, CI/CD, observability, and the tooling that helps engineers ship and operate systems with less manual work.
Lately I have been focused on AI-agent workflows: runtime controls, approval gates, MCP tools, review loops, provenance, and safer automation around real engineering work.
Based in Bern, Switzerland. Relocating to Prague, Czechia.
- Platform automation: Kubernetes/cloud platforms, GitOps, Terraform, CI/CD, golden paths, and self-service workflows.
- Data and integration platforms: Kafka, Kafka Connect, API platforms, event-driven workflows, and cloud migration patterns.
- AI-assisted engineering systems: agent workflows, MCP tools, runtime enforcement, review loops, and human-gated automation.
- Research and evaluation: practical work on tool-call safety, AI coding workflows, and failure modes in AI-assisted software delivery.
| Project | What it does |
|---|---|
| captatum | MCP web-fetch for agents: fetch any URL, render JS when needed, SSRF-guarded, with a provenance receipt on every response. Fetched content is treated as untrusted data, never instructions. |
| mcp-sso | OAuth 2.1 for remote MCP servers: spec-correct resource-server verification plus an AS bridge giving clients DCR/PKCE while your existing IdP stays the identity source. |
| qratum | Local librarian for AI coding sessions — capture, search, redact, and review Claude and Codex work. (qratum.dev) |
| spec-reviewer | Local-first Markdown spec reviewer with source-anchored feedback for agent and human review loops. |
| handoff-guard / handoff-guard-ts | Structured-output validation for LLM workflows with retry feedback. |
| pasteforward | Small CLI utility that makes image paste work in Claude Code and Codex over SSH. |
| engineering-os | The process and quality rules behind how I ship: specs, review gates, and verification loops. |
| Project | What it does |
|---|---|
| Edictum | Runtime governance for AI agents: allow, block, or require approval on tool calls, and record every decision before it executes. SDKs for Python, TypeScript, Go, and OpenClaw. |
- Mind the GAP — Text Safety Does Not Transfer to Tool-Call Safety in LLM Agents. arXiv, Feb 2026.
- gap-benchmark — Measuring text-safety vs tool-call-safety divergence across regulated domains.
- ai-monoculture — Measuring architectural convergence across 6 frontier models and 144 completions.
Roles where I can combine platform engineering, backend systems, developer tooling, and AI-assisted automation:
- Staff / Principal Platform Engineer
- Infrastructure Automation Engineer
- Developer Tools / Internal Developer Platform Engineer
- AI Platform / Agent Tooling Engineer
- Hands-on Technical Lead for platform or automation teams