Local AI for working with documents
We are developing Jerzyk, an open decision model that classifies Polish letters: it recognises document type, deadline and urgency. It can run on an organisation's own computer, without handing documents to an external provider.
Why we develop local AI models
Public offices, law firms and organisations increasingly use AI with documents. Many of them, however, need to analyse letters without handing them to an external provider, for example because of personal data, professional secrecy or internal rules.
Part of this work is simple decisions: what kind of letter is this, is a deadline running, who should handle it. These do not need a large cloud model. A small model that runs on site and answers in a fraction of a second is enough.
We work on online safety, so we also look at AI from that side: where the data goes and who processes it.
How we work
- Local. The model can run on your own computer or server. Document contents are then processed on site.
- Open. Code, model and benchmark are public under the Apache 2.0 licence. Anyone can check them and repeat our tests.
- With results and limitations. We also publish where the model is weak, and compare it with larger models on the same data.
- Polish first. We train the model on Polish letters and Polish rules on deadlines.
Jerzyk 1.0
Jerzyk: a Polish AI model for classifying letters
Recognises document type, deadline and urgency in court and administrative letters, accounting documents and customer e-mails. Runs locally or inside your own document workflow.
Example, results and limitations →Work with us
Do you work in a public office, law firm, accounting office or non-profit and want to analyse documents without sending them outside? Write to us and we will check whether it can be done locally.
Contact us