Skip to main content
Open for project inquiries

Custom AI models

Have your own data and a job you want a model to do? We work with you on post-training and pre-training, from a small 4B model to an ambitious mixture-of-experts program.

Vero beside model-building materials in a warm, tactile workshop.
Vinci illustration · A concept, not a product screenshot

Build around your use case.

Our released Vinci models show our post-training work in practice. Bring us your domain, your data, and the tasks that matter. We’ll work through the training approach, evaluation, and where the model needs to run.

A training plan

Choose a starting model or pre-training approach, organize the data, and define the compute and evaluation needed for your use case.

A model you can evaluate

Train and test against representative tasks, with results, limitations, and a comparison to the agreed baseline.

A path to deployment

Agree the model artifacts, documentation, licensing, and runtime requirements your team needs to use and maintain the result.

Tell us what you have in mind

  • What the model should do, with a few representative tasks or examples.
  • The data you have, its permitted uses, and any privacy or residency requirements.
  • Where it should run, your budget, and the scale you have in mind.

What we agree together

  • Training scope, data handling, evaluation, and ownership of the resulting work.
  • Model and dataset licenses, compute requirements, and deployment responsibilities.
  • For a trillion-parameter MoE ambition, the feasibility, data, capacity, and investment required before committing to a training program.

We build with Vinci where it fits, and choose other open-weight models and tools when they better serve your needs.

Built for your requirements

Small model or big ambition?

A compact 4B model and a trillion-parameter mixture-of-experts model need very different programs. We’ll work with you to choose a scale that makes sense for the task, data, and resources you have.

Start with your use case. We can discuss adapting an existing open-weight model or scoping a pre-training program, then agree how you’ll judge the result.

How to check AI-generated work

What should your model do?

A short introduction is enough to start. Tell us what your team needs, and we’ll work through the right approach together.

Discuss your model