Open source
I build in the open where it counts.
Most of my career has run inside private orgs, so my GitHub isn't a wall of stars. Tarka is where I've put my reliability engineering into the open — Apache 2.0, self-hosted, and free to run.
The project
Tarka — good incident triage without your best engineer.
Tarka is an open-source AI investigation agent that ingests Prometheus alerts and delivers a structured triage report in under 60 seconds — what failed, how severe, and the exact commands to run next. Built from the same reliability engineering principles I bring to every engagement.
Stop being the human runbook.
Tarka encodes the investigation steps your senior engineers carry in their heads. Whoever picks up the alert can run a real, structured triage — without waking anyone up.
A full report before Slack catches fire.
27 diagnostic checks across Prometheus, Kubernetes, and logs. What failed, how severe, ranked hypotheses, and the exact commands to run — in under 60 seconds.
Your infra. Your data. No SaaS.
Apache 2.0, fully self-hosted. Sensitive incident data never leaves your environment. Run it from a laptop CLI or wire it into Alertmanager as an automated webhook pipeline.
Why open source
The same principle I sell.
The best reliability work leaves a team stronger than it found it — the knowledge stays after the engagement ends. Tarka is that idea as software: it takes the investigation steps a senior engineer runs from memory and puts them in the open, for anyone to run.
It's also honest proof of how I think. You can read the code, see the checks, and judge the judgment — before we ever get on a call. That's worth more to me than a logo wall I can't show you.
Want the same thinking on your systems?
Tarka is a taste of how I approach reliability. If you'd like that judgment applied to your infrastructure, let's talk.