Reach VR is a virtual reality recovery platform for frozen shoulder rehabilitation, using therapist-informed assessment, adaptive exercises delivered through games, and a companion dashboard for progress tracking.
I took over the project from its prototype stage and led it to production. As the lead engineer, my work included building an API network stack to enable communication with our Node-based middle-tier, developing an VR core system compatible with controllers and hand inputs via OpenXR with passthrough, and implementing an agent system powered by LangChain on our own AWS instance. I also equipped designers with practical tools like level editors tailored for each mini game, as well as an editor for prototyping AI Agent behaviors (prompt engineering, tool configuration, testing, etc).
The Pergamos System
The Pergamos System is an AI-driven coaching platform designed to assist patients in improving shoulder health through personalized feedback and insights. As part of the core development team, I helped design and build the system from the ground up — including a FastAPI-based backend powered by LangChain and LangSmith to generate contextual responses using OpenAI models. The infrastructure was provisioned and managed via Terraform, deployed on AWS EC2 instances, and integrated with Unity for real-time user interaction. The platform also includes custom web-based UI editors for testing and refining the AI’s behavior. The system’s interactive coach, Galen, serves as the patient’s AI avatar, guiding users through exercises and providing adaptive feedback.
Mini games
At the heart of the concept are the mini games designed to make rehabilitation engaging and effective. With the help of our design team, I prototyped and developed a variety of fun and interactive experiences that ensure patients remain motivated and engaged throughout their recovery journey. Some of the challenges included simulating unrealistic but fun balloon physics without the use of Unity's built-in physics engine, stay performant in VR when using elaborate environments, solve issues of depth and scale perception, and many more.