SplitStep · a product case study by Vibhav Gupta

A phone at the fence clears its throat and says:
"Long by two feet. 78 miles per hour."

SplitStep turns the smartphones players already own into a tennis coach, counter, and line judge, so there's no hardware to buy, no cloud bill to pay, and nothing to install on the court. This page makes the case for the idea, which means the market gap it attacks, the inventions that make it possible, and why the economics are structurally different from everything else in the category.

$0hardware cost to the player

$0cloud inference cost, forever

2phones = a sensing network

1patent application drafted

27→65%real-world ball detection, own models

11days, idea → court-tested

The gap in the market

Tennis sensing today makes players choose between price tiers, not product tiers:

tier	examples	the catch
Stadium systems	Hawk-Eye, installed smart courts	$10k+ per court, someone else's court
Court hardware	net-post cameras, sensor arrays	another gadget to buy, charge, mount, and lose
Single-camera apps	vision-only phone apps	one camera guessing at physics it can't see; fails exactly when light, angle, or a net-clipped ball defeats it
SplitStep	the phones already in your bag	none

Every incumbent treats sensing as a hardware problem. The observation that SplitStep was founded on is that players already carry the hardware. A modern phone has a millisecond-accurate microphone, a fast camera, and a radio, and when two of them are coordinated, they're a small officiating system. So the missing product was never a device, but rather the orchestration layer.

The inventions

Insight 01 Two phones that only believe what they both heard

Any single microphone can be fooled by wind, by pocket rustle, or by the next court over. The creative leap was to make agreement between devices the filter, so a sound only counts when two phones that are placed apart both witness it in a physically consistent way. That means the noise doesn't get filtered out, because it never gets in at all. This idea, which treats the phones as corroborating witnesses, is the core of the drafted patent application, and it's an area where the prior-art search found open white space.

Insight 02 Sound knows when. Vision knows where. The court knows how big.

Cameras sample time too coarsely to catch a bounce, and microphones catch it to the millisecond but can't see. When you fuse them on one clock, you can measure what neither sensor can measure alone, and the court's own painted lines, which have regulation dimensions, turn four taps on a screen into full metric geometry. There are no tripods, no measuring, and no calibration hardware, because the court calibrates itself.

Insight 03 The second phone is the product, three times over

The same "spare" phone does three jobs. It's the strike microphone that unlocks serve speed and net-fault detection. It's the announcer in your pocket, so the calls are spoken where you stand and not across the court. And it's the remote viewfinder, which puts a live view of the mounted phone in your hand, so you can aim, zoom, and calibrate a camera whose screen you can't see. Competitors ship a companion app for score display, but SplitStep makes the second device a sensor, a speaker, and a control surface.

Insight 04 Follow the courts, not the roadmap

The original wedge was wall practice, until real-world testing surfaced a market fact, which is that practice walls are disappearing from Los Angeles parks. So within a day the product pivoted to serve practice, which can be tested solo because it needs no partner, which is playable on any court, and which is immediately useful. On top of that, every serve that gets recorded is training data that improves the vision models, so it's a feature that funds its own R&D.

Insight 05 Accessories yes, gadgets no

I hold a disciplined line on hardware. SplitStep will recommend a $15 fence clamp, or the Bluetooth earbuds you already own (for voice control from the baseline), but it never manufactures electronics. That keeps the margins software-shaped and avoids inventory and support, and as the freedom-to-operate research showed, it also steers around active patents that cover exactly the net-post camera gadgets that others build.

The product today

Serve PracticeYou prop a phone behind the box and serve from the far end, and you get spoken calls with distances, serve speed, net faults, and a placement map, all hands-free with voice control.

Wall & Rally ModesHit counting, streaks, personal bests, and spoken milestones, which work solo or with two phones cross-checking each other.

Own vision modelsA 1.16 MB ball detector that was trained from scratch on licensed and synthetic data, and one training run ships to Android and iPhone identically.

Works anywhereThe phones pair directly even with zero connectivity, so a court in a dead zone works exactly like one at a club.

Strategic judgment

Patent position, built before launch

An adversarial prior-art campaign across commercial, academic, and patent databases mapped the landscape before a single public disclosure. It showed that the multi-device acoustic mechanism sits in identified white space. A provisional application is drafted, and two competitors' patent families were claim-charted to confirm freedom to operate.

A cost structure incumbents can't follow down

Everything runs on the player's own devices, so there's no cloud inference and no per-user compute cost, and it's fully private by construction. The marginal cost of an additional user is effectively zero, and the product also works offline, which is something that cloud-dependent competitors structurally cannot match.

A data flywheel that respects the user

The models launch on licensed and synthetic data, so no user footage is required. But every practice session can, if the player opts in, contribute exactly the real-world imagery and acoustics that the models lack. That way usage compounds into accuracy, and accuracy compounds back into usage.

Execution leverage, managed like a team

I ran an AI engineering pair the way I'd run a high-output team, which meant clear constraints, ruthless priorities, review gates before every release, and a standing rule that nothing ships without a number or a field test. The result was 29 shipped builds, 5 model generations, a patent draft, and a competitive teardown in 11 days, all from one person's direction.

How I operate, mapped to the work I want to do

Product managementI found the wedge by testing on real courts, not in a deck. I killed my own favorite feature on market evidence ("practice walls are disappearing"), and I repointed the product in a day. I set one measurable gate for "it works" and refused vanity metrics, and that included re-measuring a competitor baseline from a claimed 83% down to its true 31 to 37% before I trusted it.

Engineering managementI ranked priorities by user pain, and I set adversarial review gates that caught release-killing bugs before the field did. I kept a same-day turnaround from field report to root cause to shipped fix. I kept scope disciplined, which meant that deferred features were named explicitly along with the evidence that would be needed to unlock them. I also managed IP risk like an engineering constraint, so the patent search and the freedom-to-operate work came before public disclosure.

Data & ML engineeringI built the data supply chain end-to-end. That covered licensed and synthetic training data, so there's zero user-data dependency at launch, and a leakage-safe held-out evaluation. It also meant one attributable change per model generation, which took real-world detection from 27% to 65%, plus operating-point tuning that halved false alarms without retraining, and a deployment pipeline that ships one set of weights to two platforms with verified numerical parity.

The vision

Serve practice is the wedge. The arc from there is two phones at opposite fences working as a synchronized officiating pair, with line calls that have honest confidence bands, serve speed, match analytics, and a coaching layer on top, which adds up to a consumer Hawk-Eye that's built from hardware the players already own. Feasibility work puts the core capabilities within a 12 to 18 month horizon, and the mechanism that makes it possible is the one that's already drafted for patent.

Where I fit

I'm looking for roles where this way of working compounds, which means product management, engineering management, or data engineering, and especially roles where a product needs someone who can hold the market story, the technical reality, and the execution cadence in one head. SplitStep is what that looks like when I'm pointed at a problem I love.