Automating the post-training loop forany business

Trusted every day by teams that build world class software

Logo 01 Logo 02 Logo 03 Logo 04 Logo 05

We’re building toward a world where shaping a model is as routine as shipping code.

We imagine a world where shaping a model to your context is as routine as shipping code. Research engineering should be accessible to every software engineer, and eventually, to everyone.

Underneath the surface diversity, all of it reduces to a single practice — signal design — the craft of communicating intent and feedback to a system that is already intelligent but not yet aligned to your problem.

The last 20% of dev effort creates the most value.

Agent deployments fail all the time in production

Hallucinations, errors, poor judgement
Systems/requirements change all the time
The way to fix and address agent issues is complex

Research engineering
delivers the recipe.

Emerging middle of the market who want to own their intelligence stack.

Open weights are now a real escape hatch.

Months behind on its benchmarks, fits on consumer hardware, and beats SOTA when post-trained for a vertical.

Open weights are now a real escape hatch.

Months behind on its benchmarks, fits on consumer hardware, and beats SOTA when post-trained for a vertical.

Open weights are now a real escape hatch.

Months behind on its benchmarks, fits on consumer hardware, and beats SOTA when post-trained for a vertical.