Technology
Engineering
precision.
On every court.
Matchlytics is built on the same standards that drive aerospace systems, where accuracy is not a target, it is the baseline.
Founder & CAIO
João Silva
Aerospace Engineer · Computer Vision Specialist
Aerospace engineering trained me to build systems where failure is not an option. I apply that same standard to every model we ship.
My background is in aerospace engineering and computer vision, fields where a 1% error rate is not a rounding issue, it is a failure mode. I carried that discipline into Matchlytics from day one.
When I first looked at what was available for padel analytics, I saw a market full of approximations packaged as insights. Ball tracking that lost the ball under a backlight. Stroke classification trained on insufficient data and fragile principles. Numbers with no methodology behind them.
We built Matchlytics to fix that. Every model is validated. Every metric is defined. Every result that reaches a player is one we can stand behind.
What accuracy actually looks like
These are the raw outputs of our models, no post-processing, no cherry-picked frames. What you see is what runs on every match.
3D Ball Trajectory
The ball's full flight path reconstructed in three dimensions from a single camera. By constraining every detection to the physics of projectile motion, we recover true height, depth and speed, the spatial picture that previously demanded a calibrated multi-camera rig.
3D Pose Lifting
Full-body skeleton reconstruction in 3D space from a single 2D camera feed. Every joint is tracked across the entire court, arms, hips, knees, giving coaching staff biomechanical insight that was previously only available in lab settings.
Player Re-Identification
When players leave the court and re-enter, the system recognises them instantly, no resets, no ID swaps. This requires deep visual encoders that build a persistent appearance model for each player, not just a bounding box.
Position Stability
Player positions in real-world coordinates stay locked even when players stand still. Without de-noise engineering, detection jitter accumulates into phantom movement, corrupting distance and heatmap data silently.
We only ship results we would stake our reputation on.
Verified before shipped
Every model is benchmarked against labelled ground-truth data from real padel matches. We do not release until numbers meet the bar.
Built for adversarial conditions
Outdoor glare, indoor fluorescent flicker, glass court reflections, players overlapping, the system is trained on the hardest cases, not the easy ones.
No black-box outputs
Every metric we surface has a traceable definition. We know exactly what we are measuring and why, and so do the players who use it.
"Most computer vision products optimise for demos. We optimise for real matches, under real conditions, with real consequences for players who trust the data."
João Silva · Aerospace Engineer · Computer Vision Specialist
See it in action