Case study
Chalk Golf is our own golf app. We designed it, built it and run it, from voice scoring to shot tracing. It's where we prove the methods we use on client projects, with real users and no signal on the 14th hole.
Golfers want to know where their shots are lost, but nobody wants to type between shots. The tools that show ball flight or swing positions usually need a launch monitor, a coach or both. Chalk puts those tools on the phone a golfer already carries, and makes them quick enough to use during a round.
What we built
Golfers hold the mic and say what happened, such as “seven iron, pulled it left, short of the green”, and the shot appears on the scorecard.
How it works: Speech recognition runs on the phone, so shots can be logged with no signal. A language model then turns each sentence into structured shot data. We found where recognition fails on a windy course, such as “tee shot” heard as “T-shirt”, and fixed those cases one by one.
Draws the flight of a golf shot over an ordinary phone video, with no launch monitor or sensors.
How it works: The app listens for the strike, finds the ball, follows it through the frames and fits a realistic flight path to the points it tracks. If it can't tell which ball is the player's, it asks for one tap rather than guessing.
Steps through the ten positions of a golf swing and picks out the one thing to fix first.
How it works: Pose estimation finds the golfer's body in each frame. From that we place the ten swing positions, draw plane lines and measure tempo, shoulder turn, hip turn and spine angle, then score the swing.
Shows where to aim a putt and how hard to hit it.
How it works: ARKit and the iPhone's LiDAR sensor measure the slope between the ball and the hole, and a physics model of a rolling ball works out the line and pace. A scanned green can be saved as a contour map and read again later.
How we approached it
Changes to the tracer and swing analysis are scored against a library of real golf clips before they ship, so a fix in one place can't quietly break another.
The tracer asks for a tap when it can't identify the ball, putt reads are labelled as estimates, and swing analysis says it reads what the camera can see, not clubhead speed.
Speech recognition runs on the device, heavier video work runs on our own servers, and paid AI calls are used only where they make a real difference.
Golf courses have poor signal. Rounds are saved on the phone first and sync when a connection comes back.
Most of the work in a product like this isn't AI. Chalk also includes:
See it for yourself at chalkgolfapp.com or on the App Store.
Start with the job a person is trying to do, test against real examples, and be honest when the system isn't sure. That works for invoices and customer emails as well as golf shots.