CV Engineer For Boxing AI Startup

About us

We're building an AI boxing coach. A fighter films a round on their phone, we tell them what's wrong with their technique, and then we build them a training plan to fix it: specific drills targeting their specific faults. The kind of coaching you'd get from a good trainer, for people who mostly train alone.

We're a remote team of 4. Early stage, shipping fast, no process overhead.

Where we are

• A working punch classifier running on pose data
• 15 form rules live in production: feet apart, hip rotation, full arm extension, guard drop, elbow flare, and others

• An in-house labeling team producing our training data

   

What you'd work on

• Adding form rules. Some are geometry and heuristics on skeleton data, some need small models trained on our labeled clips. Knowing which is which is part of the job.
• Teaching the model defensive movement: slips, rolls, blocks. The current classifier is built around punches thrown from two hands, so head movement probably needs a different approach. Working out what that approach is would be yours.
• Improving the punch classifier over time: more classes, better accuracy.

• Owning the labeling spec. You decide what the labelers annotate and make sure what comes back is usable.

   

You

• 2+ years computer vision. Pose estimation or action recognition strongly preferred.
• Comfortable with temporal models over sequence data, not only static image work.
• For at least the first couple of months you'll be the main developer, as I'll be focused on product and partnerships. We'll talk regularly and I'll set the priorities, but how things get built is yours. Nobody is going to review your approach for you, and there's no existing playbook to follow.
• We work over video, camera on. Hours are flexible, but you're reachable and you show up to calls.

What we offer

• Real ownership of the CV work. You decide how things get built, not just implement someone else's spec.
• Competitive salary, paid monthly in USD
• Fully remote
• Flexible hours. We care about what ships, not when you're online.

How we hire

Three steps, and we move fast:

1. Application. Tell us two things. First, relevant experience: action or gesture recognition, pose or skeleton data, video pipelines. What you built, and what the numbers were. Second, why this job specifically.
2. Call. 20 minutes, video. We go through one of your projects properly and you ask us whatever you want.
3. Paid test task. Real data from our dataset, scoped to around 4 hours, flat fee paid on submission whether or not we move forward.