EMG-GPT
A causal Transformer trained on residual-quantized wrist EMG, with transfer evaluation on continuous 3D hand-pose estimation.
- Model
- 40.5M params
- Signal
- 16 ch · 2 kHz
- New-user MAE
- 12.8°
Biomedical engineer · Machine learning for biosignals
Ettore Magni
I work at the intersection of machine learning, biosignals and neurotechnology — from foundation models for muscle signals to wearable EEG.
- Now
- ML Research Intern, AWEAR Technologies
- Research
- EMG-GPT, first-author paper on arXiv
- Education
- MEng Biomedical Engineering, Imperial · First Class Honours
Selected work
Machine learning, biosignals and neurotechnology.
Markerless pose feedback for RL-controlled FES
A camera-based feedback layer connecting movement tracking to a pre-existing reinforcement-learning-controlled stimulation system.
- Pose models
- 10 reviewed
- Pipeline
- ≈25 FPS
- Study
- 4 participants
Read case study
Experience
Work and research, latest first.
Jul 2026 — now
Research
Machine Learning Research Intern
AWEAR Technologies
Personalised sleep staging from single-channel wearable EEG, with cross-device validation against a clinical-grade reference.
Jun 2026 — now
Research
EMG-GPT Research Project
Foundation Models for Biosignals · with Imperial College London researchers
First-author paper on arXiv: a 40.5M-parameter autoregressive Transformer that learns transferable structure from muscle signals. Code and weights released.
Oct 2025 — Jan 2026
Industry
Data Engineer Intern
Corticale · Genova
Preprocessing, quality control and atlas mapping for 1,024-channel invasive electrophysiology recordings.
Sep 2024 — Jun 2025
Research
Markerless Pose Feedback for RL-Controlled FES
Imperial College London
Built and evaluated a camera-based movement feedback layer for a pre-existing reinforcement-learning-controlled stimulation system.
Education
Academic background and selected coursework.
2021 — 2025
Imperial College London
MEng Biomedical Engineering · First Class Honours
Relevant modules
- Reinforcement Learning for Bioengineers
- Brain Machine Interfaces
- Digital Biosignal Processing
- Principles of Biomedical Imaging
- Computational Neuroscience
- Image Processing
- Signals and Control 2
- Probability and Statistics for Bioengineering
2017 — 2021
Lycée Victor Hugo, Florence
French Baccalaureate, ESABAC
Writing
Technical articles and explanations.
PPO: how one simple innovation solved reinforcement learning’s stability challenge
From clipped objectives to a PyTorch implementation and applications in rehabilitation engineering.
Read on MediumBeyond work
Interests, languages and what I’m reading.
Mathematics, philosophy, Japanese culture and surfing.
SportSurfing; participated in three competitions.
LanguagesItalian, English and French — native or bilingual across all three.
Currently reading
Principles for Dealing with the Changing World OrderRay Dalio
MeditationsMarcus Aurelius