DNNR Tech · 2023–present

Swing Like Pro

On-device computer vision that tracks a golfer's body pose frame by frame and compares the swing with a pro's in 3D. No sensors, no uploads, free. The first app I built end to end, now in its fourth year.

0 uploadsall pose tracking runs on the phone
3Dside-by-side comparison with pro swings
Freeno private coach, no extra hardware
Role
Solo product designer and engineer
Team
Solo
Timeline
2023–present
Stack
Flutter, on-device pose estimation (TensorFlow/MediaPipe), 3D skeleton comparison, Firebase
Swing Like Pro comparing a user's swing with a pro's

The problem

The best way to fix a golf swing is a private coach, and most people can't afford one at the time and place they practice. Existing swing apps cost extra, needed sensors, or gave feedback that beginners couldn't act on.

From feasibility to product

When Google released a free pose-estimation model, I tested it in the browser first with a small JavaScript library I open-sourced, tracing the body points a swing comparison would need. It worked well enough on a single camera to justify a mobile app, since people practice at the range, not at a desk.

Pose estimation feasibility
Tracing body points and movement from two videos with pose estimation

The app records or imports a swing, tracks pose on the device, detects swing phases automatically, and overlays the user's skeleton against a pro's in 3D so you can see the difference at the top of the backswing or at impact.

Decisions along the way

  • On-device only. Video never leaves the phone. It is both the privacy promise and what makes the app free to run.
  • Fewer controls, not more. After beta feedback that there were too many toggles, I reserved the one accent color for a single primary action per screen, replaced the eight-phase labels with the two that matter to beginners (top and impact), and fixed three fundamentals as the default checklist.
  • Baselines over numbers. Guide lines for head sway and swing corridor made measurements understandable. One user wrote that the measurements were "très utiles", which was the first sign the feedback format worked.

Results

  • Live on both stores since 2023 and still improving: organic installs roughly 2.4x higher in 2026 than in March on product quality alone, with no paid acquisition.
  • Taught me the full loop of shipping a consumer app alone, which is what DNNR Tech grew out of.