Research Project
Speaktrum
Speaktrum is a machine learning-powered companion platform for identifying auditory markers associated with Parkinson's disease. Using XGBoost trained on Parkinson's patient and control datasets, the system achieves roughly 85% accuracy and is designed to support progression tracking alongside UPDRS metrics.
RoleML + Full-stack Development
TimelineOngoing
TechnologyPython, XGBoost, Flutter, Supabase, Fly.io
1 / 4
What I worked on
The scope crossed product thinking, interface design, and implementation.
01
Training and evaluating an XGBoost model on Parkinson's and control voice datasets.
02
Identifying auditory features correlated with Parkinson's disease detection and transforming datasets between media formats.
03
Optimising performance for CPU based inference.
04
Deploying the model using Fly.io containers and integrating backend services with Supabase.
05
Developing a Flutter companion app for monitoring progression alongside UPDRS-based metrics.