AI/ML · Project case study
I-THANKe — Physical Therapy Motion Tracking
A Unity-based physical therapy app that uses neural-network posture detection and gamification to help patients perform recovery exercises correctly and stay engaged.
- Focus
- AI/ML
- Timeline
- 2021
- Delivery
- Completed
- Case study
- Available
Project details
This case study covers the problem, implementation, outcomes, and practical lessons from the project. The technical narrative below is based on the project record and supporting LinkedIn export.
Skills developed
Technical and product skills applied or strengthened while delivering this work.
- Unity
- Neural Networks
- Motion Tracking
- Gamification
Project story and lessons learned
Overview
I-THANKe is a Unity-based physical therapy application that uses motion tracking and a neural network to check whether a patient is performing their therapy exercises correctly, wrapped in gamification to keep people coming back for their sessions. It was an early project for me, and the one that first pulled me toward AI and healthcare technology.
The Problem
Physical therapy only works if patients actually do their exercises, and do them correctly, often without a therapist watching every session. I wanted to explore whether motion tracking and machine learning could stand in for some of that in-person feedback, while making the sessions themselves more engaging than a printed exercise sheet.
What I Built
I built a Unity application combining guided exercise visuals with movement-based feedback, powered by a neural network model that assessed whether a user's movements matched correct therapy form. On top of the core tracking, I added gamification features to make sessions feel less like a chore, plus therapist-facing tools for tracking patient activity over time.
Key Features
- Neural network-based posture detection for real-time movement assessment.
- Guided, interactive exercise sessions built in Unity.
- Gamification mechanics designed to improve patient engagement and adherence.
- Therapist-facing data tools for tracking patient activity across sessions.
Challenges
Translating "correct form" into something a model could actually score was harder than expected — real patients don't move like a textbook diagram, and the system needed enough tolerance to be useful without being so loose it missed real mistakes. Presenting the project at three medical technology conferences also meant learning to explain a technical system to a clinical, non-technical audience.
What I Learned
I-THANKe was my first real exposure to applying AI in a healthcare context, and it's directly where my interest in that space — and later, CS-M — came from. It taught me that the hardest part of healthcare tech usually isn't the model, it's designing something patients will actually want to keep using.