Post-training shouldn't need a frontier lab.
The most capable open models are a download away. What separates a generalist from a model that's excellent at your job is post-training — and until now, doing it well took a research team, a GPU cluster and a year of tooling. POFE exists to change that.
Why we built it
We spent years post-training models at frontier labs and hyperscalers. The pattern was always the same: the training itself was the easy part. The hard parts were finding the data, trusting the evals, getting GPUs when you needed them and getting the model into production without a second team rebuilding everything.
Every company we spoke to was rebuilding the same stack from scratch, usually badly, usually twice. So we built it once, properly, and made it run anywhere — including inside the environments where the most valuable data lives.
What we believe
The eval is the product. If you can't measure it, you can't improve it, and if your eval is wrong you'll improve the wrong thing. We build evals first.
Your model should be yours. Weights, data and pipelines belong to the customer. No lock-in, no hostage adapters.
Small and specialised beats big and general. For most production tasks, a well-tuned 8B–32B model beats a frontier API on quality, latency and cost. We have the numbers.
- Founded2025, by post-training researchers and infrastructure engineers
- TeamResearch engineers from frontier AI labs, hyperscaler ML platforms and quantitative finance
- OfficesSan Francisco · London · remote across EU and US time zones
- CustomersFrom two-person startups to Fortune 100 insurers and global banks
- Backed byLeading AI-focused venture firms and operators from top AI labs
- Open sourceContributors to the major open post-training libraries
We're hiring
Research engineers (RL, evals), distributed-systems engineers and forward-deployed engineers. If you've post-trained a model you're proud of, we'd like to hear about it.
Get in touch →Bring us a model and a metric.
Tell us what you want the model to do better. A post-training engineer will come back within one business day with a plan, a compute estimate and an eval design.