How KartCast Is Built
The hardest problem in karting is not the racing
It is that almost none of it gets seen. A club race puts forty karts on track across eight hours, generates hundreds of hours of footage and thousands of lap times, and almost all of it disappears by Sunday night. KartCast exists to fix that, and the fix is a technology problem.
What runs today
Live on kartcast.com right now, at real tracks, with real drivers.
Live and on demand
Every session streamed and recorded, playable on a phone in the stands or a laptop three states away. Chapters are deep-linkable, so you can send someone the lap instead of the race.
One identity, everywhere
Sign in with a LapTime account and you are the same driver here, at the kiosk, and in the app. One racing record, owned by the driver, portable across every track that runs on it.
Built to be found
Structured data on every session and track page, human-readable URLs, and pages that render in under a second and a half. The paddock searches on Monday morning. We intend to be the answer.
What we are building
None of the below is live yet. It is what the roadmap is pointed at, and why the compute matters.
LapTime Connect
Results ingestion that works at any track, on the timing system it already runs. No rip and replace, no vendor lock, no asking a track to change the one system that cannot go down on a Saturday. Once results flow, the Monday recap writes itself and the player gains live positions, gap timers, and named drivers.
Vision that follows the kart
A camera pointed at a corner sees everything and understands nothing. We are training models to change that: detect every kart on track, read the number on the sidepod, and follow it lap after lap. That is what turns eight hours of raw footage into the ninety seconds that matter to one family, automatically, for every driver at every club race.
Every lap, worth watching
Grassroots karting generates more footage than anyone can watch and almost none of it reaches the people who care. The end state is simple to describe and hard to build: a driver finishes a session and their best laps, their pass, their mistake, are already cut, captioned with real timing data, and waiting on their profile. Not a highlight reel for the fast kid. One for everyone.
We detect karts, not people
Most of karting is under eighteen. That single fact shapes every technical decision we make about vision, and it is a decision we made early rather than one we will bolt on later.
Our models look at karts and at the numbers on them. Not faces. Not people. Kart number resolves to an entry, and an entry resolves to a driver, through the same permissioned data layer that governs every other surface in the LapTime family. A minor never appears on a public leaderboard, in a sitemap, or in structured data, and location is never more precise than the venue.
It is the harder way to build it. It is the only version we are willing to ship.
Member, NVIDIA Inception
KartCast is a member of the NVIDIA Inception program
NVIDIA Inception is a program for startups building on accelerated computing. Membership gives us access to technical training, developer tools and SDKs, preferred pricing on hardware, cloud credits through NVIDIA and its partners, and a network of engineers and investors working on the same class of problem.
What that changes for KartCast is straightforward: training vision models on thousands of hours of race footage, and running them across every session at every track, was priced out of a grassroots karting company. It is not any more. The work described above is the work the membership makes reachable.
NVIDIA Inception is a startup program. It is not a partnership, sponsorship, or endorsement, and nothing on this page is a statement by NVIDIA.
Build this with us
KartCast is a LapTime production. If you run a track, the fastest way to see where this goes is to put your races on it.