# Private AI models that run on your own hardware

## What an engagement can include

- An architecture and security plan to your standards: network isolation, authentication, logging and data retention.
- Model selection by evaluation on your tasks, with accuracy and failure cases for each model.
- Open-source serving behind one stable interface, so a model that passes the evaluation can replace the current one without changes to the tools that call it.
- Operating, update and rollback procedures, with training for your IT staff.

[How Justin evaluates models](https://justj.dev/evaluation)