We 're just ending padel beef.
We are not selling $6,000 kiosk traps. We do not do sensor-filled rackets. We just track the ball, kill score amnesia, and cancel cage beef in real time with a standard 1080p camera and net-side tablet.
a n d t h a t ’ s a l l
What actually is Tapy?
Tapy is an automated AI referee and live match dashboard running on an on-court tablet.
No sensors in rackets. No $5k hardware trap. Just a standard 1080p 60 FPS IP camera mounted on the fence, an edge computer vision pipeline, and a net-side tablet managing the match in real time.
Why amateur padel is broken
The Problem: Pure Delusion & Chaos
• Score Amnesia: Three long rallies in and nobody has any clue whether it’s 30–15 or 40–30. Players burn two minutes yelling across the net. Pure delusion.
• Cage vs. Glass Beef: The ball grazes the back wall and the drama starts instantly: “Bro, that was 100% glass, no cap!” vs. “Are you blind? That was straight cage!” Matches regularly end on the brink of physical violence.
• Hardware Gatekeeping: Legacy solutions like Wingfield or PlaySight demand $3,000 to $6,000+ for massive kiosks, complex wiring, and proprietary sensors. Local club owners want nothing to do with that.
The Solution: Zero-Sensor AI Ref
Tapy transforms any standard padel court into an intelligent arena in minutes. Just a standard 1080p camera on the fence, edge computer vision, and an on-court tablet at the net:
-
Real-Time Scoreboard (30 — 15 in <1s)
⚡ Instant Flip -
10s VAR Rewind (The Beef Canceller)
⏪ Slow-Mo Replay -
2D Court Radar Minimap
📍 FIFA / CS:GO -
Smash Radar & Clout Clips
🔥 +1000 Aura -
Pre-trained YOLOv8 & TrackNet
🧠 Zero Retraining -
WebSocket Stream to Net Tablet
📶 <50ms Latency
Designed to keep egos {checked}
It is enough to waste time yelling across the net. We built Tapy so you can focus on hitting 112 km/h viboras while computer vision handles scorekeeping, serve waist limits, and boundary calls in metric space.
Smash Radar Telemetry
Clock velocity live: “112 km/h Vibora”. Show the group chat who has the biggest arm.
Auto Vertical 9:16 Highlight Reels
Match wraps? Best rallies are automatically clipped and formatted for Instagram Stories & TikTok with ball trails.
How it’s cooked under the hood
The reason this won't crash during a 10-day build sprint: We train zero models from scratch.
-
Step 01 • Court Calibration
1080p 60 FPS IP Camera Feed
Mount any standard IP camera on the fence. 4 corner clicks allow OpenCV (
cv2.findHomography) to convert warped camera pixels into real-world court meters (10x20m metric space). -
Step 02 • Skeletons & Stance
YOLOv8-Pose (Pre-trained)
Runs 60+ FPS out-of-the-box on standard hardware. Pinpoints player skeleton keypoints, detects waist height for legal serve verification, and tracks net vs back glass positioning.
-
Step 03 • Ball & Bounce Physics
TrackNet + Kalman Filter
Pre-trained TrackNet paired with Kalman filtering. When the vertical trajectory abruptly flips direction in metric space, that’s a mathematically verified bounce frame.
-
Step 04 • Rules & Real-Time Sync
Python State Machine & WebSocket (<50ms)
Evaluates official FIP padel rules: Was the serve struck below waist? Did it bounce inside the box? Did it touch turf before glass or cage? Broadcasts instant score flip to tablet.
No $6,000 kiosk traps. Just standard {hardware}
Legacy companies like Wingfield or PlaySight demand $3,000 to $6,000+ for bulky kiosks, proprietary sensors in rackets, and cutting into concrete courts. Local club owners want nothing to do with that. Tapy runs on any standard $60 IP camera and an iPad or Android tablet at the net. Plug in, calibrate in 4 clicks, and let AI ref.
Every killer feature was designed to survive 4-player high-intensity padel warfare. We train zero models from scratch, deploy in minutes, and keep matches 100% beef-free.
-
Zero Model Retraining
Nobody is manually drawing bounding boxes around 5,000 padel balls until 4 AM. YOLOv8-Pose and TrackNet run 60+ FPS straight out-of-the-box.
-
The 10s VAR Rewind
The Beef Canceller. Whenever someone denies an out-call: tap the big red button at the net, and the last 10 seconds instantly loop in slow-mo. Argument dead, ego checked.
-
2D Court Radar (Minimap)
Top-down radar view just like FIFA or CS:GO. Four glowing dots (Volt Lime & Electric Cyan) track player positions live in metric space. Call out who camped at the glass.
-
Smash Radar & Clout Clips
Clock smash velocity live (“112 km/h Vibora”). When match wraps, Tapy automatically edits best rallies into vertical 9:16 highlight reels for TikTok & IG (+1000 Aura).
-
Real-Time Score Flip
The ball bounces twice or sails out -> the score flips automatically on the tablet in under a second (<50ms WebSocket latency). No guessing, no shouting across the net.
-
Pure Portfolio Gold
Nails the complete full-stack machine learning package: applied deep learning, geometric computer vision, low-latency WebSockets, and clean product UI.
Tapy cancels the beef
of players and club owners
-
For players
Score amnesia is permanently cured. Stop screaming across the net about 30–15 or whether the ball hit cage. Tap the 10s VAR button, check the verified slow-motion bounce, see your 112 km/h smash velocity, and walk off court with receipts and +1000 aura.
-
For Club owners
No $6,000 kiosk gatekeeping. No drilling concrete or tearing up turf for proprietary cables. A single $60 IP camera and a tablet turns any court into a viral smart court that members will queue to book. Zero maintenance headaches.
Problems that we solve
Score Amnesia Eliminated
Three rallies in and everybody forgot the score? Tapy tracks every bounce in real time and flips the score on the net-side tablet in under a second. No more 2-minute arguments or delusional guesses.
Cage vs. Glass Beef Cancelled
“Bro that was 100% glass no cap!” vs “Are you blind? Straight cage!” Tap the big red button at the net: the last 10 seconds instantly loop on the tablet in slow-mo. Argument dead, ego checked, friendship saved.
Zero $6,000 Hardware Extortion
Legacy systems charge $3k to $6k for bulky metal kiosks and sensors inside rackets. Tapy uses standard off-the-shelf 1080p IP cameras and tablets. Affordable for any local padel club.
Automated Social Clout
When match wraps, get automated vertical 9:16 highlight reels ready to flex on IG Stories & TikTok with live smash telemetry and glowing ball trails.
10-Day Role roadmap
The reason this won't crash during a 10-day build sprint: We train zero models from scratch. Here's how we ship to Demo Day:
{Vision Pipeline}
Camera calibration with OpenCV Homography (4 corner clicks) + pre-trained YOLOv8-Pose for player skeletons at 60 FPS.
{Ball Tracking & VAR}
TrackNet & Kalman filtering for bounce frame detection + rolling video buffer for instant slow-mo replays.
{Rules & WebSockets}
Python state machine for official FIP rules (serve waist limit, turf before glass) + low-latency WebSocket broadcaster (<50ms).
{Demo Day Flex}
Tablet dashboard UI + 2D court radar minimap with Volt Lime & Cyan dots + vertical 9:16 highlight clip exporter.
FAQ
Do I need special sensors in my racket or ball?
What hardware is required to run Tapy?
How does Tapy know if a ball hit the glass or cage?
Why don’t you train models from scratch?
Can I get my match highlights for TikTok & IG Stories?
What is the 60-Second Demo Day Pitch Script?
“Padel is the fastest growing racket sport on earth with 25 million players, but actual matches are pure chaos. Three rallies in, nobody has any clue whether it’s 30–15 or 40–30. Players yell across the net for 2 minutes, and then the ball grazes the back wall: ‘Bro that was 100% glass, no cap!’ vs. ‘Are you blind? Straight cage!’ Matches end on the brink of violence.”
“Meet Tapy: the automated AI referee and live match dashboard running on an on-court tablet. No $6,000 kiosk extortion. No sensors in rackets. Just one standard 1080p 60 FPS IP camera on the fence, an edge computer vision pipeline, and a net-side tablet managing the match in real time.”
“Under the hood: 4 clicks calibrate the 10x20m court via OpenCV Homography. Pre-trained YOLOv8-Pose tracks player skeletons at 60 FPS. TrackNet and Kalman filtering pinpoint verified bounce frames. A clean Python state machine enforces official FIP rules with under 50ms WebSocket latency. We train zero models from scratch.”
“The killer feature? The 10s VAR Rewind button at the net: tap it, and the last 10 seconds instantly loop in slow-mo. Argument dead, ego checked, friendship saved. Plus live 2D court radar and automated 9:16 highlight reels for TikTok. That’s Tapy — zero beef on court.”
Teacher Defense: How does OpenCV Homography eliminate wide-angle parallax?
cv2.findHomography solves the 3×3 perspective transformation matrix H that maps camera image space to real-world metric space (X, Y) ∈ [0, 20]m × [0, 10]m. Because the court floor is a planar surface, homography provides exact metric coordinates for ball bounce locations and player heatmap positioning regardless of camera mounting angle or lens distortion.