ML Researcher & Engineer · MSc Computer Science @ Sapienza Università di Roma (May 2026)
I work at the intersection of research and production — from publishing at IEEE to deploying ML APIs on AWS. My current focus is continual learning and vision transformers for my thesis, while building end-to-end ML systems on the side.
- Continual Learning for Computer Vision (Thesis, ongoing) — Evaluating pretrained ViTs under TIL/DIL/CIL protocols, structured review of 60+ papers on catastrophic forgetting
- ML for Air Pollution Prediction — IEEE ViTECoN-2023 · Survey of 6 ML models across 6 NAAQS pollutants · 6 citations
DocDelta — Semantic Document Diff API
Detects meaning-level changes between document versions, not just character diffs. Classifies changes as added, removed, or modified (numerical vs. wording).
FastAPI Sentence Transformers Docker AWS ECS Fargate CloudWatch
~840ms warm latency · CPU-only inference · Stateless REST design
LossyTextCompressor — Semantic Embedding Quantization
Compresses text by quantizing GTR-T5 embeddings and reconstructing via Vec2Text beam search. Reveals a sharp quality phase transition — performance collapses beyond 8-bit.
GTR-T5 Vec2Text Quantization ROUGE-L BLEU PyTorch
6-bit optimal: cosine similarity 0.964, ROUGE-L 0.776
CNN-based yoga pose classification from live webcam input. Frame predictions smoothed across consecutive frames for stability.
CNN TensorFlow/Keras JavaScript Real-time Inference
ML & Research: PyTorch · TensorFlow · Transformers · CNNs · NLP · Continual Learning · Model Evaluation
Engineering: FastAPI · Docker · AWS ECS/Fargate · CloudWatch · REST APIs
Languages: Python · Bash · C++ · Linux
