I’m Matteo Fasulo, an open-source enthusiast and AI Engineer, passionate about building intelligent systems that bridge research and real-world applications.
Between my BSc in Computer Science and MSc in Artificial Intelligence (currently at the University of Bologna, with Master's thesis at ETH Zurich), I’ve worked on projects covering:
- Generative AI & LLMs - Prompt-based evaluation, in-context learning, and retrieval-augmented generation. Mainly conducted at Outlier AI.
- NLP (CLEF 2025) - Enhancing Transformer-Based Embeddings with Sentiment for Subjectivity Detection in News Articles. Part of CLEF CheckThat! 2025 Lab on Subjectivity Detection. https://github.com/MatteoFasulo/clef2025-checkthat
- NLP (EXIST 2023) - EXIST 2023 task in finetuning models for Sexism Detection in Social Media and EDOS task from SemEval 2023 using In-Context Learning with zero-shot and few-shot learning. https://github.com/MatteoFasulo/Sexism-detection
- Biosignal Modeling - Pretraining of a Foundation Model for EMG signals with ViT-based architectures using Self-Supervised Learning (masked autoreconstruction). Several downstream tasks were then tested, namely EMG gesture classification, EMG-to-pose prediction, and Silent Speech Recognition. Part of the Master's thesis at ETH Zurich.
- Data Mining - Amazon review sentiment analysis using R Shiny, ggplot, Plotly, and text mining pipelines.
- Cloud Computing - Scalable WebApp using Docker Swarm, Ansible, ShinyProxy, Traefik, InfluxDB, and Grafana monitoring. https://github.com/MatteoFasulo/ScalableWebApp
- Embedded AI - Low-level optimization and implementation on Xilinx/Versal VCK190. https://github.com/MatteoFasulo/Versal-C-Code
Some experience (and occasionally, contribution) worth to mention over open source projects:
- OpenAI Whisper under HuggingFace transformer.js with ONNX Runtime and WebGPU.
- Adversarial Robustness Toolbox (ART) hosted by Linux Foundation AI & Data Foundation (LF AI & Data).
- pgmpy library for Bayesian Networks.
- Whisper-TikTok – A side project I started purely for fun: an AI-driven tool with Microsoft Edge TTS, OpenAI Whisper, and FFmpeg to automatically generate TikTok-style videos—with transcriptions, voiceovers, and background content—just by tweaking a simple JSON file.
