Lausanne, Vaud, Switzerland
HackaHealth is an association organizing hackathons aiming at engineering personalized solutions for people with motor disabilities to improve their daily living. Technical Leader in user-centric projects. • Strong involvement within the association with 1 year as President. • Technical leader for diverse software project. Project 1 (4 months duration): Assistive keyboard for stroke survivor. Hand2keyPressedRehab provides a solution to press keys on keyboard based on the detection of residual hand gestures (Leap-motion) for stroke survivor. Project 2 (4 months duration): Remote mouse control with fingertip tracking. HandTracking emulates a mouse based on hand tracking. The technology is based on Computer Vision(HSVFilter, HandPose estimation).
• Led the development of machine learning pipeline for detecting speech disorders by integrating signal processing and deep learning methods such as MFCCs, ASR models, and ResNet. Collaborated closely with clinical experts to establish performance metrics and experimental protocol. Presently, collecting data for model validation before deployment for patient. • Contributed to research focused on the potential use of large language models (LLMs) for clinical purposes. Retraining of BERT on Electronic Health Records to enable the prediction of risk stratification for chronic kidney disease. Publication in process.
• Developed and deployed a risk score to assess risk of cardiovascular disease based on survival modeling using UKBiobank dataset with explainability methodologies for health-related recommendations. The algorithm was further tested on large scale in US population and will be later release in US pharmacies. • Created an extensive framework for the collection, cleaning, and analysis of large datasets, with production-level quality code. Proficiently deployed and operationalized machine learning models using VertexAI and Docker on Google Cloud platform.
• Development of machine learning algorithms able to predict movement intention in healthy people and patients using brain signals • Real-time deployment of brain controlled devices for rehabilitation or assistive technology (e.g. exoskeleton, wheelchair) • Feature Engineering and Hyperparameter Tuning in time and frequency domains • Signal Processing of brain and physiological signals
• Teaching Machine Learning to more than 100 students with practical exercises on neural datasets • Supervising group of students, teaching them how to analyze brain signals and design a Brain-Machine Interface
Nicolelis lab, Duke University, Durham, North Carolina, USA Title: Speech-like communication using invasive Brain-Machine Interface • Analysis of neural data recorded in monkeys with invasive electrodes • Pipeline for neural data preprocessing with Unit Testing • Design of a machine learning model predicting monkey’s behavior (object selection, cursor movement) - GUI-interfaces for preprocessing and analysis of neural dataset