The project is being developed by the following Sekcja Sztucznej Inteligencji members: Eldar Aliekpierov, Natalia Błaszczyk, Maciej Cichoń, Agata Świetlik.
The solution is aimed towards helping TUL staff and aspiring students navigate the university's documentation. The user can ask the bot a question related to some document from the TUL database (e.g. "How many terms are there to pass an exam?") and the bot will respond with accurate, up-to-date information on the topic (e.g. "Students have 3 tries at passing an exam..."). Something more relevant for future TUL students is the "course recommendation system", with it being one of the features of the bot. A high school graduate (maturzysta) can send the bot their exam grades, subject preferences along with other information that could help determine the best-fit course for the individual, and the bot will respond with its recommendation based on the provided information.
Сurrently houses a minimal implementation of RAG, along with the beginings of not-so-minimal one. Different base models are being tested on TruthfulQA, with different embedding models being tested on LEPISZCZE and KLEJ. The project is still under heavy development, implying possible functionality, UI or any other kinds of changes.
docker run -p 6333:6333 -v ~/path/to/loca/vector/database:/qdrant/storage:z qdrant/qdrant
ollama create newmodelname --file path/to/model/file
uvicorn app:app --reload