Backend and platform engineer with 10+ years across distributed services, data platforms, and cloud-native delivery on Kubernetes. I turn unsettled requirements and tangled system boundaries into clear structures that teams can build on — and lately, into workflows where AI coding agents do the same.

I build tools that let coding agents do real engineering work reliably — and I run them that way every day.

Most of what I write now lives in Sympoies: open-source CLIs, runtime policies, and dashboards for teams of AI coding agents (Claude Code, Codex, DeepSeek Harness, local models) working side by side with humans. The tools are extracted from a private setup where several agents work in parallel under a coordinator, with code review, test-first gates, release gates, and hands-on acceptance before anything counts as done.

Agent tooling and runtime

• nils-cli — a Rust workspace of focused CLIs with automation-first JSON contracts: Git helpers (git-cli, git-scope), forge automation (forge-cli for GitHub/GitLab PRs and issues), agent session management (agent-session), validated commits (semantic-commit), API testing (api-rest, api-gql, api-grpc, api-websocket), evidence and devlog tools, and more. Installable from the Homebrew tap.
• agent-runtime-kit — one shared source of truth for agent skills, hooks, and policies across Codex, Claude, and Hermes.
• dsh-runtime-kit, dsh-workbench, dsh-applications, dsh-plugins — runtime policies, a Web/TUI workbench, and plugins for DeepSeek Harness.
• agent-sandbox — a pinned Linux container that validates an agent toolchain combination before it reaches real hosts.

Visibility for humans

• symphony-board — a read-only, cross-provider board for GitHub/GitLab issues, PRs/MRs, activity, and typed relationships.
• host-monitor — a central, read-only dashboard for Linux host resources, services, containers, probes, and journal errors.

Developer environment

• zsh-kit — a modular, self-contained Zsh environment.
• night-owl-cli — a synchronized Night Owl theme for the whole terminal.
• nils-alfredworkflow — Alfred workflows for macOS.

• Backend and platform architecture — service boundaries, API and data-model design, and shared backend baselines that keep many projects consistent.
• Data platforms — Spark, ELK, PostgreSQL, Elasticsearch, and Neo4j pipelines and query systems at billions-of-records scale.
• Multi-agent engineering workflows — dispatch, review lanes, merge queues, acceptance, handoff, and recovery when sessions stall or get compacted.
• CLI and contract design — small composable tools with stable JSON output, typed errors, and deterministic exit codes that both people and agents can rely on.
• Languages — Rust, TypeScript / Node.js, Go, Java / Spring Boot, Python, Scala, and shell.
• Cloud-native delivery and CI — Docker, Kubernetes, AWS, GCP; GitHub Actions and GitLab CI; test-first gates, scoped CI, nightly and release validation, Homebrew distribution.
• Self-hosted infrastructure — running services, monitoring, and local LLM inference on a small fleet of Linux and macOS hosts.

Evidence over claims: an agent saying "done" is a claim; a passing test, a reviewed diff, and a hands-on check are evidence. Tools should make the right thing the default and explain how to fix the wrong thing.

Interested in agent tooling or want to compare notes? Open an issue or a discussion on any Sympoies repo.