The student will create a pipeline in Jupyter Notebook that will allow the user to pre-train and fine-tune Large Language Models on private data locally. The input to the pipeline will be user private data (typically software documentation or internal regulations). The pipeline transforms the data into a form suitable for fine-tuning, e.g., a set of question-answer pairs. The final fine-tuned model will then be able to serve as a chatbot for queries related to private data.

The student will apply the pipeline to the UltraGrid project documentation (https://www.ultragrid.cz/), test more models created with this pipeline and evaluate their usability. Manually created fine-tuning data will also be included in testing.