The promise sounds like marketing: select a game clip and get a short, pose-based movement breakdown without sending the video to a server. It is fair to be skeptical. So let's open the hood. Here is exactly what Rally's AI coach does with your video, step by step, and — just as importantly — what it does not pretend to do.
Step one: it finds a visible player pose
The moment your clip is selected, Rally runs pose estimation entirely on your device. At each sampled moment it selects the largest visible pose and locates body landmarks such as wrists, elbows, shoulders, hips, knees, and ankles. Ball and paddle tracking are the natural next step and are on the roadmap: a pickleball is small, fast, and spends a lot of its life as a blur, which makes it the hard part.
This is why framing matters. Rally needs a clear head-to-toe view so it can follow the same visible player pose across the clip. Prop the phone against the fence, keep your full body in frame, and avoid crowds crossing in front of the camera. No special camera, wearable, or markers are required.
Step two: it turns pose landmarks into movement metrics
Pose landmarks become a small, allowlisted set of numeric measurements: ready-posture percentage, movement-rhythm events, lateral range, arm-extension events, analyzed duration, sample count, and detection quality. Rally does not currently identify serves, returns, dinks, shot outcomes, or ball position.
The current analyzer measures visible body movement. It does not grade your swing, follow the ball, or decide whether a shot was successful.
That boundary matters. Pose-only measurements can show whether your stance stayed ready or how much lateral ground you covered, but they cannot tell whether a drive won the rally. Ball, paddle, and shot-outcome tracking remain roadmap capabilities.
Step three: it returns a bounded breakdown
Rally maps those measured values through fixed thresholds into a short breakdown: a headline, strengths, focused fixes, and a drill. The live analyzer can discuss ready posture, movement rhythm, coverage, and arm extension; it cannot claim late contact, paddle timing, third-shot quality, or reset height from pose alone.
Crucially, it stays inside those measured signals. If the clip does not contain enough clear pose samples, Rally says so and asks for a better angle instead of inventing feedback. When the signal is usable, the result stays short, specific, and yours.
What it will not do
Rally's coach is honest about its limits, and you should be too. It will not replace a great in-person pro who can feel the timing of your swing and put a hand on your shoulder to fix your grip. It does not see intention, nerves, or the conversation you had with your partner between points. What it does is make the time between lessons count — so you show up to your next session already working on the right thing.
- It runs on your device, so raw clips never leave your browser.
- It measures pose-based movement, not the ball, paddle, or shot outcome.
- It hands you a short, ranked list — not a data dump you'll ignore.
That is the entire trick. No magic, no hand-waving — just bounded pose measurement, honest labeling, and a short drill-oriented breakdown. Prop the phone against the fence and try it; if you save the result, Rally stores the numeric analysis while the raw clip stays in your browser.