Computer vision, simulation and machine learning, deployed in the tools your coaches already open. A system built to measure, fitted to how you work.
Tennis
Cycling
Press
SailingTennisCyclingPressSailing
Tracking and reconstruction built on your own footage.
Fluid, structural and bioenergetic models that answer in seconds.
GPS, IMU, video and timing, landing where coaches already work.
Pre-match briefs and post-race teardowns in your staff’s format.
Every squad holds different data, and every question arrives in its own form. We start from yours and shape the work around it.
With your staff, on site or remotely. The question, the data that exists, and the outcome we commit to.
Where the data doesn't exist yet, we build the capture around the question.
Models built on your data, tested against what your staff already knows.
In the tools your coaches already open. Source, documentation and training included.
Ball and both players tracked in 3D from the match feed. Every shot measured, then analysed.
With Mouratoglou Academy
Expected threat with pressure and line height in the model, from tracking data.
What we build in footballOne broadcast file and a roster in. Every player tracked and named on a court plan, with possession and shots.
Pipeline built and published
Whole races rebuilt from the boats’ own tracking, and simulation fast enough to answer between runs.
With SailGP Australia
Rider and team position tracked from the race broadcast. Nothing mounted on the bike.
With Soudal Quick-Step
Our football pipeline extended to rugby: detection and tracking calibrated onto the pitch, then transition threat and expected points decomposed into what actually decided the title.
Study published April 2026
The same method as a squad mandate, run for a single athlete: training and race files turned into a physiological reading, and the athlete's own account of each session collected and kept alongside the data.
Triathlon with Marjolaine Pierré and Clément Mignon · rowing with Tim Dumain
Your sport is not there ? Let's add it to this list !
L'Équipe asked what would have happened with a round bar. We rebuilt the trajectory from the 1976 television images and ran 100,000 simulations.
Shot quality never changed; conversion at the rim fell from 80% to 17%. A 25-page report built on our own tracking of the broadcast.
Court homography, identity rebuilt from jersey numbers and roster, possession and shots aligned to the official play-by-play.
Six Nations 2026Most tries scored and most conceded in the same championship. Transition threat at 0.48 per turnover against a 0.27 mean.
Spain always plants one extra body in the central corridor, and France builds through a defender and two forwards rather than its midfield.
Bayern had the ball, PSG had the dangerous spaces. Four mechanisms, and the physical cost that almost broke PSG after sixty minutes.
Engineers and researchers who have competed themselves and worked inside teams. The people who build your system are the people your staff talks to.
No. Every engagement is a bespoke build scoped around one team and one outcome. You end up owning the system, source included.
Most follow a competition cycle, a season for football and cycling, a championship window for sailing and triathlon. Discovery is one week; a first deployed output lands early in the cycle.
No. Roughly half our clients start with footage, spreadsheets and GPS exports. Part of the mandate is building the collection pipeline properly the first time.
Thirty minutes on the question your staff cannot answer yet. We say on the call whether it is tractable.
NDA before discovery, on requestDelivered with Mouratoglou Academy.
A tracker for the ball and both players in 3D, built on the academy's own multi-camera feeds, so that every shot is measured and then analysed.
Thirty minutes with an engineer. We say on the call whether we are the right people.
Tracking built on a broadcast feed, turned into expected threat, pressing and line height.
An expected-threat model that adds defensive pressure and line height from tracking data, built for opponent preparation and in-match substitutions.
Thirty minutes with an engineer. We say on the call whether we are the right people.
One broadcast file and a roster in, every player tracked and named on a court plan.
A tracking chain built across four competitions and stress-tested on Olympic and NBA footage. A court homography turns pixels into positions in metres. Identity is rebuilt after tracking, from jersey numbers, the roster and the rules of the game, so the output names all ten players on court. It runs in about the duration of the broadcast.
Thirty minutes with an engineer. We say on the call whether we are the right people.
Delivered with SailGP Australia.
Neural surrogates trained on high-fidelity CFD of the F50 foiling catamaran bring simulation latency from hours down to the regatta window. On IMOCA, dense sail-shape reconstruction from sparse on-board cameras enables in-situ trim optimisation.
Thirty minutes. We say on the call whether we are the right people.
Delivered with Soudal Quick-Step.
Drag coefficients for riders in breakaways and leadouts, estimated from monocular broadcast video and calibrated against wind-tunnel runs. Nothing mounted on the bike.
Thirty minutes with an engineer. We say on the call whether we are the right people.
Our tracking pipeline, extended from football to rugby for the 2025 season.
France scored the most tries in the championship and conceded the most. We reverse-engineered how a team with the worst defensive record still won it. Transition efficiency converted broken play at roughly twice the rate of the rest of the competition, the goal kicking left nothing on the field, and one winger moved the scoring baseline by more than a try a match.
Thirty minutes with an engineer. We say on the call whether we are the right people.
The squad method, run for one person. Delivered with professional triathletes and rower.
One report among those we build, shown to give an idea of the form. A six-step lactate test fixes where the thresholds sit. A heart-rate model fitted on the athlete’s own training predicts the cardiac cost of a session and shows which conditions move that cost most. What a report contains follows the question asked.
Training and race files turned into a bioenergetic profile, then into a pacing plan for a specific course and forecast.
Training files arrive on their own. After a session a chatbot asks the athlete how it went, in their own words, and the answer is stored alongside the data rather than lost in a message thread.
Every week the report reads both together: training load against recovery, aerobic durability, sleep, heart-rate variability and resting heart rate against the athlete's own rolling baseline, and the next target race.
Where the numbers and the athlete's account disagree, the report says so, and it flags a sensor that has stopped reporting before the discrepancy is mistaken for fatigue.
The plan stays with the coach. The report puts the week in front of them, and says where it is uncertain.
Quantum Sport Analytics was founded in 2025 in Paris. A small team of engineers and researchers, building bespoke data and AI products for clubs, academies and rights holders. We work from our own studio and go on site when the question calls for it.
Marin GautierCo-founder
Hugo StublerCo-founder
Idriss Ben M'radLead data scientist
Mathieu BouteillerData scientistA small team building bespoke products for elite staff.
Every engagement produces something your staff owns: a model, a pipeline, an interface, yours to run and to extend. We work from our own studio and come to you when the question needs the pitch, the boat or the training ground.
Every output is made for a coach who has to decide now.
A drag model built for a WorldTour leadout sharpens a foiling catamaran. Each sport we work in sharpens the next.
Research is part of the work. Methods are published before they reach a season, and the code we write stays with the team that paid for it.
The 12 May 1976 European Cup final, rebuilt in 3D from the television images alone, then run 100,000 times against a round bar. Published as a magazine feature and a podcast.
Method write-ups, reports and notes from inside the squads.
12 May 1976, European Cup final in Glasgow. Santini's header comes back off the square bar and Saint-Étienne lose 1–0 to Bayern. From the 1976 television images alone we rebuilt the ball's trajectory in 3D, then ran 100,000 simulations against a round bar: a 26% chance of a goal. One in four.
San Antonio led by 29 at half-time and lost game 4 by one. Tracking every player every second shows the shots never got harder — shot quality held flat while shot-making collapsed, and conversion within four feet of the rim fell from 80% to 17%. The report follows the chain: Towns pulls Wembanyama out of the paint, the pick-and-roll behind him goes from 0.70 to 1.82 points a possession, Anunoby locks Fox, and a nine-man rotation runs out of legs against twelve.
The published pipeline: a court homography turning pixels into metres, an identity layer rebuilding exactly ten named players from jersey numbers and the roster, possession and shot detection, and a clock reader aligned to the official play-by-play.
France scored 30 tries and conceded 19, both tournament records, and took the title with a 50-metre penalty on the last kick of the championship. The study decomposes the paradox: transition threat at 0.48 per turnover against a 0.27 mean, 27 of 27 place kicks, and one winger with nine tries in five matches. Match-by-match, spatial and attribution analysis.
Third straight defeat, same script each time. Rebuilt from our computer-vision pipelines: Spain always plants one extra body in the central corridor, and France builds through a defender and two forwards rather than through its midfield.
Top assist provider of the tournament, involved in 16% of France's key passes and shots for half that share of the minutes, 11.4 km a match. We rebuilt his profile from raw data and put it next to the great orchestrators.
Bayern had the ball, PSG had the dangerous spaces. Four mechanisms stood out, and after sixty minutes the physical cost of adapting almost broke PSG.
A full pipeline that reads every key event of a match straight from video: passes, recoveries, expected goals, tactical shifts. Applied to the 2025 final.
A demo app for tennis where the staff choose which overlays and analytics they want to see, built around the details that decide matches: break points, first serves, endurance in long rallies.
Nothing in this category yet.
Tell us the question your staff can't answer yet.
We hire engineers who want to build. You own a pipeline end to end, on real sports data, and your work reaches a staff that uses it inside the season.
Build the AI models and the pipelines around them, and lead a team of data scientists on mid-to-large projects. Master's in applied mathematics or computer science, strong Python, solid grounding in modelling, linear algebra and differential equations.
Six months contributing to our machine-learning and computer-vision models applied to sport. For an M2 student in applied mathematics or data science with a real appetite for sport.
Send a CV, a short letter and any relevant work or project. The hiring post on LinkedIn.
We read spontaneous applications. Tell us what you would build, for which sport, and send a CV with anything you have shipped.
The link may be old, or mistyped. Everything on the site is one click away below.
The site is hosted by Cloudflare, Inc., 101 Townsend Street, San Francisco, CA 94107, United States. Telephone: +1 650 319 8930. Site: www.cloudflare.com.
The structure of this site, its texts, its graphics and the models presented on it belong to Quantum Sport Analytics unless stated otherwise. Any reproduction or representation, in whole or in part, without prior written permission is prohibited.
The video sequences shown illustrate analysis work carried out on television broadcasts. The rights to these images belong to their respective holders and are cited here as a technical illustration. The trade marks and logos of the organisations mentioned belong to their owners.
The processing of data collected through this site is described in our privacy policy.
This document covers the data collected on this site. Work delivered to a club or an athlete is covered by a separate agreement.
Quantum Sport Analytics, a SAS with its registered office at 29 avenue Le Nôtre, 92330 Sceaux. For any question about your data: marin.gautier@quantum-sport.ai.
Nothing is collected without your knowledge. The site sets no advertising cookie and no third-party tracker, and does not follow you from site to site. That is why you are never asked to accept a consent banner.
Your data is accessible to the partners of Quantum Sport Analytics. It passes through our host and through the service that carries the forms, both acting as processors on our behalf. Your data is never sold, rented or passed to third parties for advertising.
You have the right of access, rectification, erasure, restriction, objection and portability, and the right to withdraw your consent at any time. Write to marin.gautier@quantum-sport.ai. If our answer does not satisfy you, you may refer the matter to the CNIL, 3 place de Fontenoy, 75007 Paris, cnil.fr.
This policy may change. The date below marks the latest version.
Data processed under a mandate, including the health data of an athlete we follow, falls outside this policy. It is covered by a separate contractual agreement with the organisation or athlete concerned, and by an information notice given to the people followed.
Last updated: 9 September 2026
Thirty minutes to put the question your staff cannot answer yet. We will say on the call whether it is tractable in this cycle, and what data it would take.