Know where to check.
An answer should lead to a document and a specific passage.
We’re building a language model to support work with Polish law. With a specific source, the relevant date and room for human judgment.
In development · working name: lexpl
Project presentation film · archival v4.
An answer should lead to a document and a specific passage.
Law changes. The version of a provision is part of the question.
AI supports the work. People retain judgment and responsibility.
A legal question rarely ends with a single sentence. We are designing a tool that helps move from a situation to evidence you can inspect.
A question, a document and the relevant date. The starting point is your work.
The goal is to identify the relevant provision and version, with a passage to read.
The source should be available beside the answer. Insufficient evidence should be made clear.
We are developing our own tokenizer and model weights. Our work covers retrieval, answers and quality evaluation. lexpl remains a working name.
We do not present the project as a finished legal adviser. Research results need context and separate assessment.
A Polish-language corpus and our own tokenizer.
Pilot models and source-retrieval research.
Tests, error analysis and source verification.
Conversations about real tasks and success criteria.
Finding provisions, reviewing documents and questions that require checking the legal basis.
Document workflows, internal procedures and preparing evidence for decisions.
Collaboration on evaluation, pilots and the next stage of technology development.
Which provision applies
to this situation?
Illustration of the intended workflow. This is not a model response.
Watch the project film or explore the presentation about AI in organisations. These materials illustrate the direction and approach.
Open the presentationWe are looking for partners to help assess usefulness in real work. Tell us which documents, questions and processes matter to you.
Discuss a collaboration