Physicist turned ML engineer, currently finishing my Master's in Computer Science at the University of Trento (expected Spring 2027). I work on equivariant machine learning. My thesis applies this to quantum machine learning, but the core interest is equivariance itself, across classical and quantum deep learning. I'm drawn to problems where deep mathematical structure does real work.
🔎 Looking for roles in:
- Equivariant / geometric deep learning
- Machine learning & deep learning engineering / R&D
- Software engineering
📍 Based in Italy — open to roles in Veneto or remote.
| Degree | Field | Institution | Year |
|---|---|---|---|
| M.S. (in progress) | Computer Science | University of Trento | Expected Spring 2027 |
| B.S. | Physics | Midwestern State University | 2022 |
Master's Thesis (in progress) — Equivariant Models for Quantum Machine Learning: Image Denoising
Designing group-equivariant architectures for image denoising, where the symmetry is enforced by construction rather than learned from augmented data.
Bachelor's Thesis — Higher-Order Corrections to Bhabha Scattering
Computed radiative corrections to electron-positron scattering in QED.
A C₄-equivariant model for image classification that is invariant to 90° rotations by construction. The symmetry is built into the architecture through group twirling, parameters shared across rotation orbits, and an invariant readout, instead of being learned from data. It is implemented as a variational quantum circuit and reproduces the published results of San Sebastian Sein et al. (test F1 0.64 ± 0.02 vs 0.63 reported) across five seeds.
equivariant·group-symmetry·quantum-machine-learning·PennyLane
A hybrid few-shot classifier on a frozen CLIP ViT-B/16 backbone, combining CoOp prompt tuning with a training-free proto-adapter and a confidence-based switch between them. On Flowers102 (10-shot, base/novel split) it reaches 91.5% base accuracy (+20.1 pts over zero-shot CLIP) and a +8.3 pt harmonic-mean gain, with hyperparameters chosen by a two-stage search on validation and results reported over three seeds.
few-shot-learning·CLIP·prompt-tuning·computer-vision·PyTorch
A from-scratch PyTorch implementation of Show, Attend and Tell — VGG16 encoder, Bahdanau-style soft attention, and gated LSTM decoder — with HDF5 feature caching, doubly-stochastic attention regularization, and a full training and evaluation pipeline on Flickr8k.
image-captioning·attention·seq2seq·PyTorch
A CLI tool for registering and recognizing faces in photos using facial embeddings stored in PostgreSQL with pgvector. The Rust binary handles CLI and database logic; a Python script generates ArcFace embeddings via DeepFace, and matches are found by cosine-distance vector search.
rust·postgresql·pgvector·face-recognition·embeddings
Core ML
Foundation Models
Quantum ML
Experiment Tracking
Languages, Data & Tools
Physics background. ML focus. Always building.