Health-tech diagnostics platform
ParkinsonAI
A full-stack diagnostic application with real-time PyTorch inference, disciplined delivery and a jury special award.
- Period
- July to August 2025
- Team
- 4-person team, T3 Foundation AI & Technology Academy
- Role
- Full-stack AI application development
- Technologies
Key contributions
- Developed a FastAPI backend, JavaScript frontend and real-time PyTorch inference layer.
- Containerized the application with Docker and deployed through an automated CI/CD pipeline.
- Received a Jury Special Award among 128 teams at the AI Bootcamp Datathon / Hackathon.
Problem
The project explored how a machine-learning diagnostic workflow could be delivered as a usable full-stack application rather than an isolated model notebook.
Context and responsibility
ParkinsonAI was developed by a four-person team during the T3 Foundation AI & Technology Academy AI Bootcamp Datathon / Hackathon.
Architecture
A JavaScript frontend called a FastAPI backend connected to a real-time PyTorch inference layer. Docker standardized the runtime, while an automated pipeline provided the CI/CD path.
Important decisions
- Keep model inference behind an explicit API boundary.
- Package the system in Docker so development and deployment use a consistent environment.
- Treat delivery automation as part of the project, not an afterthought.
Evaluation, safety and limitations
The award and team result are verified. Dataset composition, clinical validation, model metrics, latency and medical-safety controls are not publicly documented. The portfolio therefore does not present the system as a clinically validated product.
Outcome
The project received a Jury Special Award among 128 teams.