Athletique
Mobile app · Phase 2 preview

Eight sports.
One pipeline.

The Athletique iOS app captures every rep on-device. Apple Vision finds the ball, the body, and the form. CoreML scores it before the kid hits the bench. The clip and the verdict land in the family portal before they're off the floor.

On-device only — no cloud round-tripiPhone & iPad · iOS 17+Auto-edits a highlight clip
9:41
9:41
9:41
The pipeline

Capture, classify, summarize.
All on-device.

No frames leave the phone unless a parent opts in to a shared clip. The capture loop is Apple Vision's pre-trained models — `VNDetectTrajectoriesRequest` for ball arcs, `VNDetectHumanBodyPoseRequest` for body pose — wired into a ring buffer so we can look back at the release frame after a trajectory confirms.

  1. step 01
    Capture
    AVCaptureSession · 60fps default, 240fps slow-mo on Pro models. Buffer is teed to a live analyzer and a recorded clip.
  2. step 02
    Detect
    VNDetectTrajectoriesRequest + VNDetectHumanBodyPoseRequest run per-frame on a serial Vision queue.
  3. step 03
    Aggregate
    A 3-second ring buffer holds every joint position + ball point. When a trajectory confirms, we look back for the release frame.
  4. step 04
    Score
    Per-sport CoreML classifier reads pose + trajectory and emits the event (shot_made, kill, ace, …) with confidence.
  5. step 05
    Summarize
    Auto-edit the 8-second clip around the apex; render the verdict overlay; upload only on parent opt-in.
  6. step 06
    Sync
    Events land on the same `class_sessions` row the kiosk wrote. The family portal shelf updates in seconds.
The lineup

Eight sports at launch.

Each sport gets a tuned capture loop, a per-sport CoreML classifier, and a screen that reads as part of the same family. More sports queue behind these — request yours from the contact form.

9:41
Sport 01 · Basketball

The shot, scored at the rim.

Trajectory, release form, made or missed. 50 free throws becomes a labelled dataset for the season.

Apple framework
VNDetectTrajectoriesRequest + VNDetectHumanBodyPoseRequest
Events emitted
shot_attempt · shot_outcome · form_snapshot
9:41
Sport 02 · Baseball

A swing, broken down.

Bat speed, attack angle, contact point. Spray chart updates with every cage rep.

Apple framework
VNDetectHumanBodyPoseRequest + custom CoreML
Events emitted
swing_attempt · contact_event · spray_outcome
9:41
Sport 03 · Soccer

Not a shot — a sequence.

GPS + visual tracking. Heatmap, distance covered, sprint count, decision quality.

Apple framework
Core Location + VNDetectHumanRectanglesRequest
Events emitted
sprint · touch · pass_attempt · shot_attempt
9:41
Sport 04 · Tennis

Spin, depth, every stroke.

Trajectory plus pose. Forehand, backhand, serve — classified, plotted, and rallied with.

Apple framework
VNDetectTrajectoriesRequest + pose
Events emitted
serve · groundstroke · volley · ace
9:41
Sport 05 · Volleyball

Six players, one read.

Set, spike, dig — recognized per-rotation. Plus jump height from on-device pose, no force plate needed.

Apple framework
VNDetectTrajectoriesRequest + multi-pose
Events emitted
set · spike · dig · block · ace
9:41
Sport 06 · Golf

Tempo. Path. Face.

Down-the-line capture, slow-mo on iPhone Pro. Plane angle, hip rotation, club-face contact.

Apple framework
VNDetectHumanBodyPoseRequest + slow-mo capture
Events emitted
swing · address · impact · follow_through
9:41
Sport 07 · Boxing

The combo, counted.

Jab–cross–hook recognized from pose. Punch count, rate, output by round.

Apple framework
VNDetectHumanBodyPoseRequest + sequence classifier
Events emitted
jab · cross · hook · uppercut · slip
9:41
Sport 08 · Gymnastics

The line, judged.

Joint angles at apex. Beam, bars, floor — extension, alignment, rotation count.

Apple framework
VNDetectHumanBodyPoseRequest + form classifier
Events emitted
mount · skill · release · dismount
iPhone · iPad · iPad Pro

Same app, three contexts. The kid's phone for individual capture, the coach's iPad on a tripod for team practice, the iPad Pro at the rink for game footage.

Neural Engine, not the cloud

Apple's ANE handles real-time pose + trajectory in the milliseconds it takes Wi-Fi to handshake. Frames never leave the phone — the highlight clip is what gets uploaded, and only on opt-in.

Built for minors

Verification photos retained only as long as attendance reconciliation requires. Highlight clips are family-controlled — parents pick who can see what.

Coach view

The iPad on the tripod sees what the coach sees.

Mount the iPad once per court. The coach taps the rim or the endline to calibrate — the app remembers per court ID. From then on, every shot, swing, sprint, or stroke is captured, classified, and added to that athlete's session row in the shared Neon DB.

  • Single-tap rim / hoop / endline calibration
  • Per-court ID stored locally, re-prompt on drift
  • Manual tag overlay for ambiguous events
  • TestFlight beta for partner facilities first
  • Apple Sign-In · ships in month 2
  • Bring-your-own iPad — no proprietary hardware
Highlight queue
Maya R. · Gymnastics
Cartwheel · ext 9.2
12s ago
ready
Owen R. · Soccer
Left-wing cross · 47 mph
1m ago
ready
Aarav B. · Boxing
Jab-cross-hook · 0.6s
2m ago
ready
Yuki P. · Gymnastics
Beam V-sit · 8s hold
4m ago
ready
Wren C. · Tennis
Topspin BH · 1,920 RPM
6m ago
ready
Phase 2 · 2026

Tell us your sport.
We'll build the loop.

These eight ship together. Anything else takes a sport-specific capture spec and a few weeks of labelled data. We'll start from your sport on request — partner facilities get the work for free in exchange for the dataset.