Provenance is information about the history of a documented object. Provenance is represented as an annotated graph, where nodes of the graph represent entities, activities, and agents, and edges represent their mutual relations. This thesis aims to implement provenance generation for machine learning pipelines stored in ML Fflow. In particular, the student will: 1) get familiar with the current standards for provenance representation (W3C PROV, Common Provenance Model) and related tools; 2) get familiar with the use case for which provenance generation will be implemented; 3) design and implement provenance generation for the use case. The machine learning pipelines are applied in digital pathology, where the models are trained to detect carcinoma cells in high-resolution images of the human prostate. The generated provenance will enable traceability of precursors used to train the model -- for a trained model, it will be possible to find each image used for the training/testing/evaluation of the model. The resulting architecture for provenance generation will enable easy integration with the existing system.