TrackMist

See every rep. Fix every mistake.

Comment

Inspiration

Most athletes train without a coach watching every rep. Good coaching and motion-capture systems are expensive, so small mistakes in posture and technique go unnoticed until they turn into bad habits or injuries. We wanted to give any athlete a free "digital sports twin" that watches their movement, remembers their progress, and explains what to fix, using only a phone or laptop.

What it does

TrackMist is a browser-based sports analytics platform.

  • AI coach: upload a photo or video in one of 20 sports. TrackMist detects the body with pose estimation, draws a skeleton over each frame, and measures knee bend, elbow angle, torso lean, left/right balance and how much of the body is visible. It turns these into a technique score, plain-language mistakes and a drill for each one.
  • Progress comparison: each result is compared with the athlete's last session in the same sport, so they can see what improved and what got worse.
  • Voice coach: feedback is read aloud, with a choice of voice and speed.
  • Live device data: athletes can connect a Bluetooth heart-rate band or watch, or scan a QR code to stream heart rate and movement from their phone. It appears on the dashboard automatically.
  • Dashboard and streaks: technique trends, sessions per day, live heart rate and a streak tracker where the user chooses days or months and how many.
  • Everything on the dashboard comes from real sessions and devices. There is no placeholder data.

How we built it

  • Front end in HTML, CSS and JavaScript with a black and green design and a 3D rotating hero.
  • MediaPipe Pose Landmarker runs in the browser to find 33 body points, and we calculate joint angles from them.
  • Charts are drawn with the Canvas API in green shades.
  • Web Bluetooth reads the standard heart-rate service from wearables.
  • PeerJS (WebRTC) and a QR code link a phone to the dashboard without our own server.
  • The Web Speech API provides the voice coach.
  • Session history and device readings are stored in the browser (localStorage), so the athlete's data stays on their own device.

Challenges we ran into

  • Our first prototype showed illustrative numbers, which looked good but was not honest. We replaced them with measurements from real pose detection and removed all sample data.
  • Browsers limit what a web page can do. Web Bluetooth works in Chrome and Edge on desktop and Android but not on iPhone Safari, and pairing a phone by QR needs the site to be hosted online.
  • Without a backend, we had to send data between devices peer-to-peer and keep history on the device.
  • Angles measured from one camera view are estimates, so we added a visibility check that warns when the whole body is not in frame.

Accomplishments that we're proud of

  • Real in-browser pose analysis across 20 sports with no installation and no special hardware.
  • Feedback that comes from measured angles and includes a drill for every problem.
  • A working device connection through Bluetooth or a phone QR code, with a streak tracker the user can set to days or months.
  • A voice coach that makes the feedback usable while training.

What we learned

  • How pose landmarks turn into joint angles and why camera angle and framing matter so much.
  • How to use the Web Bluetooth, WebRTC and Web Speech APIs, and where their limits are.
  • That honest, clearly labelled estimates are more useful to athletes than impressive-looking numbers.

What's next for TrackMist

  • Sport-specific scoring rules built with coaches, instead of general posture checks.
  • Side-by-side playback of two sessions with skeletons overlaid.
  • Hindi and Marathi support, and asking the coach questions by voice.
  • Athlete profiles with goals, cloud sync across devices and shareable PDF reports.
  • Connecting more wearables and training-load data such as speed, cadence and workload.

Built With

Share this project:

Updates

Submission history