Start a Combat-Sports Computer-Vision API and SDK

1. 1

   Build the vision models that are hard to replicate

   The core asset is accurate, real-time punch classification, pose and keypoint estimation, and technique scoring from ordinary camera footage. Use established tools for pose tracking plus your own trained models, and invest in the labeled data and edge cases (angles, lighting, body types) that make accuracy reliable. Accuracy on messy real-world video is the whole product, and it is where most consumer apps are weak.

2. 2

   Package it as an API and SDK

   Wrap the models in a clean hosted API and a mobile SDK so a developer can add punch tracking and scoring in days instead of building computer vision for a year. Documentation, sample apps, predictable latency, and clear pricing are the product for a developer audience. You are selling time and reliability to teams who do not want to become computer-vision experts.

3. 3

   Manage the compute economics

   Vision inference runs on GPUs, so cloud compute is your main variable cost and can decide whether the business is profitable. Optimize models, support on-device inference where possible to cut cloud cost, and price usage so margins hold as volume grows. The whole crowded app field depends on this compute layer, and controlling its cost is your defensibility.

4. 4

   Sell to builders, not end users

   Your customers are AI boxing app makers, gym-platform companies, equipment brands adding smart features, and even broadcasters, not individual boxers. Sell developer relations and reliability: SDKs, docs, support, and case studies. Landing a few platform customers who embed your API creates recurring, high-margin usage revenue and makes you infrastructure they depend on.

5. 5

   Expand beyond boxing

   Punch tracking is the beachhead, but the same pose and movement infrastructure extends to other combat sports and to fitness and movement categories that currently lean on wearables. A camera-only vision layer can replace sensor hardware across many activities, so the platform can grow well past boxing while keeping the technical moat you built. That optionality is the upside of owning the infrastructure layer.

Do not take our word for the money. Put your own numbers in and see what this idea has to earn before it works. Free, instant, no account needed.

Best for: Machine-learning engineers and technical founders who want to own infrastructure, not an app

Not for: Builders who want a consumer app, or who lack ML depth and compute. The durable position is infrastructure, the arms dealer, not another app.