Biomedical Engineer · Clinical AI

I build and evaluate AI systems for clinical oncology.

I'm a biomedical engineer with an MSc from Politecnico di Milano and two years in product management. My thesis at the Istituto Nazionale dei Tumori tested whether LLM agents can be trusted to recommend lung cancer treatment. The work led to an abstract presented at the ESMO Congress 2026.

See the research Download CV

Open to PhD positions and industry roles in clinical AI. Milan, Italy

MSc, Politecnico di Milano Biomedical Engineering · 2026

Latest

Oct 2026 Our abstract on evaluating clinical AI agents for first-line NSCLC immunotherapy was accepted and presented at the ESMO Congress 2026.
Oct 2026 Defended my MSc thesis, “Evaluating Clinical Agentic AI Systems in Oncology”, at Politecnico di Milano.
Feb 2026 Joined AI-On-Lab at Fondazione IRCCS Istituto Nazionale dei Tumori, Milan, as an agentic AI research intern.

Featured research

Can we trust an AI agent with a cancer treatment decision?

How it works

MSc thesis · ESMO 2026

Evaluating Clinical Agentic AI Systems in Oncology

AI agents can already write a treatment plan for a lung cancer patient. The harder question is whether their reasoning holds up. My thesis built nsclc-eval, a framework that scores both the reasoning and the final recommendation of LLM agents choosing between immunotherapy alone (IO) and immunotherapy plus chemotherapy (IOCT) for advanced NSCLC.

Python
llama.cpp
Docker
Pydantic
Open-weight LLMs
LLM-as-a-judge

5 open-weight LLMs

25 judge × agent pairs

454 evaluations

8 metrics in 2 tiers

Selected work

Projects

All projects

Team project2025

Decoding visual stimuli from fMRI

Predicted the category of a viewed image from activity in 4,466 visual-cortex voxels, comparing MLP, BiLSTM, Transformer, 3D-CNN and BiGRU models. Handled strong class imbalance with class weights and PCA, and mapped Integrated Gradients attributions back to single voxels.

PyTorch
fMRI
Explainability

Code

Course project2025

Lung nodule malignancy classification from CT

Fine-tuned ConvNeXt-Tiny on full CT slices and on nodule crops from the LIDC dataset, for benign/malignant and 5-level malignancy. Used MixUp and CutMix against class imbalance and Grad-CAM++ for explanations.

ConvNeXt
CT imaging
Grad-CAM++

Article

Working on clinical AI?

Open to PhD positions and industry roles in clinical AI. I'd be glad to hear from you.

Get in touch