RangeGuide

After a joint injury, patients are often told to “watch if it improves,” but that improvement is hard to measure at home. One day might feel better, another might feel worse, and by the time they see a doctor or physical therapist, most of the data is based on memory. RangeGuide turns recovery into a simple guided routine. A warm AI clinician avatar walks the patient through the same movements each session, measures their range of motion, compares injured and reference sides, records discomfort, and shows whether mobility is actually improving over time. The goal is not to diagnose an injury. The goal is to create a clearer recovery timeline so a real clinician can review objective movement trends instead of relying only on scattered self-reporting.

RangeGuide works as an end-to-end demo. A patient can: Start a guided joint screening session. Choose a joint, such as wrist, elbow, shoulder, or knee. Follow a Tavus-style AI clinician avatar through movement instructions. Compare injured-side range of motion against a reference side. Record discomfort level during the session. Generate a doctor-ready report with max angle, discomfort, session timeline, and a plain-language summary. View past sessions in a history page to see whether range of motion is improving over time. The key feature is the trend view: instead of treating the screening as a one-time check, RangeGuide shows recovery as a timeline.

The app is built in React, TypeScript, Vite, and Tailwind, with typed mock data powering the full experience. We used Lovable to build a polished, navigable product demo quickly, PAL Maker/Tavus for the AI-guided clinician concept, and Magic Canvas to shape the user flow and product experience. We also used MediaPipe to measure the angles of the joints. Under the hood, the exam runs like a guided state machine. Each step has an instruction, mock movement data, angle readings, discomfort values, and transcript text. The video stage uses animated mock pose landmarks to simulate how camera-based joint tracking would work. The angle system is structured around the real geometry: three pose landmarks define a joint angle, and those readings become a session timeline. The report and history pages are rendered from typed objects, so the product already has the structure needed for real pose-estimation data later. External systems like Tavus, webcam access, pose estimation, PDF export, and clinician sharing are separated behind service stubs, so the demo stays safe and mock-first while still showing the intended architecture.

Next, we would connect the mock pose layer to real pose estimation, add proper camera calibration, and test accuracy across different body positions, lighting, and devices. We would also build a clinician dashboard where doctors or physical therapists could review trends across sessions, flag stalled recovery, and adjust rehab plans. Longer term, RangeGuide could support more joints, personalized rehab protocols, secure clinician sharing, and integration with electronic health records. The most important next step would be clinical validation: making sure the measurements are reliable enough to support real medical decision-making while keeping the AI clearly in a support role, not a diagnostic one.