Opportunities

We're building the most capable real-time sports perception system ever deployed on a mobile device.

 

Coach Ex Machina is applying that technology to create an AI tennis coach that works from nothing more than a smartphone. We're looking for our founding engineer to own the system that gives our AI coach perception.

 

The Mission: Bring Eyes to AI

Our AI Coach needs to understand what is happening on the court. This is far beyond pose detection. You're building the perception layer of an AI system: the part that decides what actually happened, how confident it is, and interprets the scene before an LLM coach ever speaks.

 

Can it detect a tennis swing? Classify the stroke? Identify the important moments? Track multiple players? Understand spacing, infer racket grip, footwork, movement, velocity, and changes in direction? Can it handle drills, partial swings, even read player body language?

 

Your mission is to build the perception engine that enables our AI Coach to see and understand the game. It must recognize not only players and strokes, but also drills, equipment, and the surrounding context. From that, it must infer what the player is trying to accomplish. The structured information your system produces becomes the eyes of the AI—the evidence an LLM coach relies on to deliver intelligent, personalized coaching.

 

All of this must operate accurately on modern smartphones, in real time and completely offline, because responsive coaching shouldn't depend on network connectivity or cloud latency.

 

Problems you'll own

Some of what we're working on has no known solution on a phone:

• Detecting and classifying a tennis stroke within a fraction of a second from a live camera stream that can never look at future frames.

• Telling strokes apart when the discriminating signal is barely present in 2D pose.

• Making models generalize to players they've never seen.

• Inferring what is happening from the entire scene—not just body pose. Distinguishing practice drills from match play, recognizing equipment and court context, and interpreting the cues that change what the AI should say.

• Turning thousands of player recordings into a data flywheel through labeling, evaluation, and continuous model improvement.

• Determining when the visual evidence is unreliable because the scene itself is ambiguous, incomplete, or physically implausible.

• Making the system honest enough to say "I don't know" rather than coach the wrong thing confidently.

 

Where this grows

Perception here isn't just about detection. The evidence your system produces feeds directly into an LLM coaching brain — multimodal fusion, real-time voice coaching, and the question of when the AI should trust the model, the frames, or neither. If you want to grow from classical CV rigor into real-time multimodal LLM systems, this role is that path.

 

You

You've built production ML or CV systems and can prove they worked: rigorous evaluation, honest holdouts, noisy real-world labels. You have real-time instincts — you know what it means that a live system can't see the future. Experience with human pose estimation, temporal video understanding, or mobile deployment is valuable; sports analytics, biomechanics, or motion analysis is a plus. If you're a tennis player, bonus! We don't expect the whole stack on day one: Swift, CoreML, and tennis are all learnable here.

 

The Role

This is a senior individual contributor role, and a founding one. You own the perception engine end to end: what it detects, what it may assert, how it's measured, and how its data improves. You'll make sound technical decisions independently and collaborate with the rest of the engineering team with minimal supervision.

 

The Team

You'll be joining an experienced team that has shipped successful products before. We have a shipping product, an ambitious roadmap, and customers we listen to carefully. We're looking for exceptional people who can push on the open problems, not just execute the known ones.

 

The Opportunity

As with any startup, you'll wear many hats, move quickly, and have enormous influence over the product. We're starting with tennis, but what you build — perception that produces calibrated evidence an AI can act on — applies far beyond a single sport.

 

As a founding engineer, compensation is weighted toward equity, although a modest salary is available. We're fully remote and flexible about location and time zone.

 

If building the future of AI perception — and solving some of the hardest real-time computer vision problems on mobile — sounds exciting, we'd love to hear from you.

​

Life's too short to work on boring problems!