“With his critical thinking and principled engineering approach, we cleaned up a mess and built a sophisticated AI coaching platform from the ground up.”
Dristix Labs Client testimonial
Enterprise AI
Enterprise AI at Hansel.
Specialized, fine-tuned models replacing general LLM calls where they are faster, cheaper, and more accurate for an enterprise AI product.
My own product · open source · Linux developer preview
Find who decided what, and hear them say it.
Overtalk is a meeting notebook for your own computer. No bot joins the call, and with its local models the recording, transcript and notes never leave the device. It handles the work around a meeting, not just the notes:
Every decision, action and open question links to the moment it was said, so you can play it back before you rely on it.
Meetings file themselves into projects, and about an hour before the next one you get a brief: what was decided, what is still open and who is coming.
Follow-ups for Linear, Jira and Confluence are drafted for you and sent only after you approve them.
Claude Code, Codex or Cursor can search your meetings and quote their sources through its MCP server.
Free and MIT-licensed. I rebuilt it from an open-source recorder into a native Rust app with a sandboxed speech engine.
I build the whole product with you, one milestone at a time. To prove one part first, start with a smaller engagement.
Starting prices exclude VAT. We agree on scope before work begins.
Build your AI product
One technical owner for the whole product: the data, the custom model, the web app and API, deployment and monitoring, and a handoff your team can own.
Within six weeks of the start, a first working version: your model running on your own data, behind an app your team can try.
You pay per milestone, and you can stop after any one and keep everything delivered.
When a general model is the right choice, the hard part is making it reliable. A working assistant on your own data and systems, with tools, guardrails, and an evaluation set, in staging. A voice mode is optional.
Find out whether your footage and data can support the product. I review the data, test the riskiest component, and map out the architecture and next steps.
