← Leveling Golf ES

How the AI Golf Caddie Works

From tee to scorecard in 5 steps. No phone. No sensors. Just your Apple Watch.

1

Start Your Round

Open Leveling Golf on your Apple Watch and tap Start Round. GPS identifies which course you're on automatically from 30 pre-loaded courses in Dominican Republic, recognized by the Federación Dominicana de Golf (FEDOGOLF).

From this moment, you don't need to touch your phone or any screen. The Watch handles everything.

GPS course matching uses geofencing with a 200m radius around each course's first tee.

2

Swing Detection Activates

Your Apple Watch's accelerometer samples motion data at up to 100Hz and the 3-axis gyroscope captures rotational movement. Together, they create a detailed motion profile of everything your wrist does.

A Core ML model running directly on the Watch analyzes this data in real time. It looks for the unique motion signature of a golf swing: the backswing acceleration, the transition pause, the downswing force, and the follow-through deceleration.

When a swing is detected, a subtle haptic tap confirms it on your wrist. False positive rate: less than 5% per round.

The model uses a sliding window buffer (MotionBuffer) and adaptive sampling that switches to 10Hz idle mode after 5 seconds of quiet to preserve battery.

3

AI Identifies Your Club

Every club produces a different swing pattern. A driver swing is long, fast, and flat. A 7-iron is shorter with a steeper angle of attack. A wedge is compact and aggressive. A putter is slow and pendulum-like.

A second ML model (FeatureExtractor + ClubPredictor) analyzes swing characteristics: tempo, arc length, peak acceleration, force curve, and GPS distance from the previous shot. It identifies which club you used with 80%+ accuracy.

When confidence drops below 65%, the Watch asks you to confirm with a quick tap. Every correction feeds back into the model, improving predictions for your unique swing over time.

Inference runs on-device via Core ML in under 100ms. The model assembles an MLFeatureVector from swing features + GPS distance + player context.

4

GPS Tracks Your Shot

GPS records your position at the moment of each swing and at your next position when you prepare to hit again. The app calculates shot distance, plots the shot on the course map, and adds it to your scorecard.

Adaptive GPS sampling adjusts frequency based on activity: higher sampling while walking between shots, lower while stationary.

Course snapping aligns GPS coordinates to known hole layouts using pre-loaded course geometry.

5

Review Your Round

After 18 holes, you have a complete scorecard with:

  • Shot-by-shot data for every hole
  • Distance stats per club
  • A visual GPS map of every shot
  • Fairways hit, greens in regulation, putts per hole

Everything syncs to your iPhone and the cloud automatically. Review your round, share scores, and track improvement over time.

Offline-first sync: data stored locally first via SyncQueue, synced when connectivity is available with exponential backoff retry.

The Technical Pipeline

Watch Sensors MotionBuffer (sliding window)
  
SwingDetector (Core ML) swing detected? haptic tap
  
FeatureExtractor MLFeatureVector (swing + GPS + context)
  
ClubPredictor (Core ML, <100ms) club ID + confidence
  
GPS + CourseSnapping shot distance + position
  
Scorecard SyncQueue iPhone Cloud

Key Technical Specs

Your Watch already has the hardware. We built the intelligence.

Join the Waitlist