Nikola J.

- Designed and led the development of vision foundation model-based mammography classification/risk prediction - Led on-premise deployment of mammography AI API - Supporting interaction with client data storage and web application.

- Co-led design and development of a first radiologist-visual-Turing-test passing diffusion model ensemble (MAMBO) for high-resolution mammogram generation - Applied scaled parameter efficient extended self-supervised learning (SSL) pretraining across 6 A100 GPUs using PyTorch FSDP to adapt DINOv2 backbone for medical use case - Developed SetTransformer-based flow-matching in SLL representation space for synthesizing sets of embeddings to augment data for training of models with weak supervision - Successfully pitched PoC ideas for securing funds.

Designed and deployed a dockerized 3D CNN-based instance segmentation inference pipeline