Independent engineering lab
Building the tools behind agentic engineering.
Open-source software, practical experiments, and engineering notes on AI platforms, coding agents, and developer workflows.
- Open-source tooling
- AI platform engineering
- Technical writing
Featured engineering work
Tools built to be used, not demoed.
Long-running work on agent tooling and AI platform architecture. Open-source projects ship from this lab; platform engagements are published as case studies within their confidentiality boundaries.
code-factorio/platform — AI platform engineering
Flagship engagement: architecture, boundaries, and trade-offs of an internal AI platform — model gateway, agent runtime, evaluation harness. Published as a structured case study; client and implementation details stay private.
Beskar — beskar.ventrion.dev
Skill management for AI agents. One curated, portable library; profiles per project; fingerprint-based change tracking; deploy into any workspace.
Rohrpost — rohrpost.ventrion.dev
Git-native ticket system for coding agents. Tickets live in the repository, every command speaks JSON, and the ready queue respects dependencies.
Recent technical writing
Notes from the workbench.
Beskar
Managing agent skills across projects without copy-paste
Why skills drift when every project keeps its own copies, and how a single library with profiles and fingerprints fixes it.
Rohrpost
Why I built a git-native issue tracker for coding agents
Hosted trackers split an agent's context between the repo and the ticket system. Moving tickets into git changes the workflow.
Platform
Building an AI platform: architecture, boundaries, and trade-offs
What an internal AI platform should own, what it must not, and where the trade-offs bite — without violating client confidentiality.
Engineering approach
Decisions over defaults.
Every tool on this site started as a concrete problem in daily engineering work: skills drifting between projects, agents locked out of the ticket queue, platform teams drowning in one-off integrations. The writing covers the decisions, the alternatives that were rejected, and what broke along the way.
Problems first
Start from a constraint that hurts in real work, not from a technique looking for a use case.
Architecture decisions
Boundaries, data flow, and failure modes — documented with diagrams and the alternatives considered.
Trade-offs
Portability against performance, control against flexibility. Name the cost of every choice.
Implementation & constraints
Real terminal sessions, real config, real-world limits — offline-first, zero-daemon, stdlib-only where it counts.
# A ticket's life, entirely inside the repository
$ rp init
$ rp new "Fix token refresh race" --type bug
$ rp ready --json
$ rp claim 1a2f
$ rp close 1a2f --reason "done"Consulting
Scoped engineering help, when you need it.
Three concrete areas, delivered as fixed-scope engagements: AI architecture reviews, custom MCP servers and integrations, and agentic engineering workflows. The tools on this site are the reference implementations.
AI architecture reviews
A structured review of your AI stack — boundaries, failure modes, cost and latency profile — with prioritized findings.
Custom integrations
MCP servers, connectors, and agent tooling that fit your systems instead of fighting them.
Agentic workflows
Designing coding-agent workflows that hold up in production: queues, guardrails, evaluation.
Prefer a concrete starting point over an open-ended engagement? So do I. One email is enough to scope it.
Engagement formatsAbout the lab
One engineer, shipping in public.
Ventrion is the independent engineering practice of a systems and platform engineer working on developer tooling and applied AI. Tools get built here, decisions get written down here, and everything that can be open, is.
- Focus
- Agentic engineering, AI platforms, developer tools
- Ships as
- Open-source CLIs, case studies, articles
- More
- About Ventrion