projects
Open-source tools and frameworks developed during my research.
NAIM
Not Another Imputation Method — A transformer-based framework for handling missing values in tabular datasets. NAIM learns cross-feature attention patterns from available information only, thus ignoring missing data.
- Stack: Python, PyTorch
- Paper: AI Open (2026)
- Code: cosbidev/NAIM
MARIA
Multimodal Attention Resilient to Incomplete datA — A multimodal transformer that fuses clinical tabular data and medical imaging for diagnosis and prognosis, handling arbitrarily missing modalities at both training and inference time.
- Stack: Python, PyTorch
- Paper: Computers in Biology and Medicine (2025)
- Code: cosbidev/MARIA
Text2CT
Text-to-CT Generation — A 3D latent diffusion model with contrastive vision-language pretraining that generates synthetic CT volumes from free-text radiology descriptions.
- Stack: Python, PyTorch, HuggingFace
- Paper: arXiv (2025)
- Code: cosbidev/Text2CT