AI at universities — with local models for research data

Knowledge assistant for research output and documentation

Knowledge and documents

Problem: Research output and documentation are scattered across faculties and repositories

How it works: A central RAG knowledge base with OCR and metadata filtering; natural-language search with source preview

What you gain: Researchers and administrators find research output and documentation with one question, with the source indicated — without searching several repositories

Student assistant for regulations and procedures

Knowledge and documents

Problem: Handling students’ questions about regulations and procedures overloads dean’s offices

How it works: A workspace with a dedicated assistant and a base of regulations; integration with ePUAP and e-Doręczenia (e-Delivery)

What you gain: Students get an answer from the regulations at any time, and the dean’s office deals with matters that require a decision

Comparing interpretations from several models

Quality and standardisation

Problem: Research work requires comparing interpretations from several models

How it works: The model council and a split view; hot-swapping the model during a session without losing context

What you gain: Interpretations from several models in one view, with the option to change the model without repeating the context

Local AI for protected research data

Security and sovereignty

Problem: Research data covered by agreements and grants must not leave the university

How it works: Routing only to local models; offline mode with local models and integrations

What you gain: The terms of grant agreements are met technically — research data stays in the university’s infrastructure

Separate AI environments for university units

Security and sovereignty

Problem: Each faculty needs its own separate environment within one infrastructure

How it works: A multi-tenant model with environment isolation; ABAC per organisational unit

What you gain: One infrastructure, one maintenance team, and each faculty has its own rules for access to data and models

Open documentation of AI involvement in research

Compliance and oversight

Problem: Research integrity requires open disclosure of AI involvement in writing

How it works: AI Disclosure Injection and a model use log as evidence of what was generated and when

What you gain: AI involvement in a research text is labelled and documented automatically — in line with the requirements of publishers and ethics committees

AI governance and clear responsibility at the university

Compliance and oversight

Problem: AI deployment at the university is blocked by the lack of a policy and unclear responsibility between faculties

How it works: An AI use policy and an agent register with an owner per unit (CDF-F2); AI Champions in faculties and the CDF Academy (F3)

What you gain: The university has one policy, known owners and AI champions in faculties — deployment no longer depends on the enthusiasm of individuals

Can a researcher enter unpublished results into an AI tool?

Only into a tool that runs in the university’s infrastructure or guarantees that the data is not stored or used for training. PROXY:AI lets you use local models, including Bielik and PLLuM

How should AI in teaching be regulated?

Through a university policy that sets out the disclosure of AI involvement in student work, grading rules and approved tools

What AI tools should researchers get?

A shared gateway to local and cloud models with budgets for faculties and units, and for administrative staff — an assistant that answers from internal regulations with a citation