Sole designer of the first PLO solver in a browser, from research to design system, still live four years on with 120+ paying subscribers
PLO Genius is a cloud-based Pot-Limit Omaha solver and GTO trainer, built to run in a browser on a subscription a player could justify.
Deepsolver (NLH) had already proven that a neural-net cloud solver could work. PLO Genius brought the same approach to Omaha: a more complex game with far fewer learning tools.
I was the sole designer, and that constraint shaped every decision. There was no onboarding team and no brand to fall back on, so whatever the product needed to teach had to be designed into the product itself.
Research, UX, UI, the marketing site and the design system. The core judgment was treating the interface as the product: the engine was proven, the players who needed it most could not read its output, so the whole design existed to make that matrix readable.
Before it, studying PLO with a solver meant buying a $5,000+ PC to run MonkerSolver. There was no affordable, browser-based alternative, so most PLO players studied without a solver at all.
A solver answers with a matrix of frequencies and stops there. The players who needed one most were the least able to read it: the tool that was supposed to teach demanded a language the learner did not have yet.
Put a PLO solver in a browser at a price a player could justify. The bar was the alternative: hardware, software, minutes per calculation. Most players would not pay any of it, so they studied without a solver.
Deepsolver had already shown a neural-net cloud solver worked for No-Limit Hold’em, so the engine was never the open question. The interface was. The product had to exist to make the solver’s output readable.
I was designing a learning tool for a game I didn't play. The usual sources failed in useful ways: beginners could not say what they needed because they did not know the game yet, and pros operated on intuition that would not translate into interface decisions. Neither could validate a judgment about how a solver should teach.
The bridge turned out to be poker stables: organizations where a knowledgeable lead managed groups of players at different levels. Those leads understood both the theory and the learning process, which made them the most useful collaborators for validating design decisions.
Validation ran through those stables. Each new surface went to a lead who both understood the theory and taught it, and the design changed on what they said: which charts confused, which drills held, where a screen asked more than a player could know yet.
The design broke the matrix into three surfaces, each answering one question a player actually asks. Each one carries its own piece of the teaching, with no onboarding team to explain anything outside the screen.
Preflop: "what should I play here." Range charts and matrices across stack sizes, positions and rake, drawn from the trained network.
Postflop: "how does my hand do against that range on this board." Equity visualisations turn the matrix into a picture.
GTO Trainer: up to four tables at once, tuned to feel like a real session so learned play transfers to it. No custom bets, so the drill stays on correct play.
The marketing site had to teach the same idea to people who had never used a solver: it led with the player's question, not with the engine, because nobody buys a matrix they cannot read.
Launch. The product found paying users, and the screens above are the ones it shipped with.
The product shipped with paying users and is still live four years later, while the engine it presented is licensed to other platforms. Both revenue lines ran through the same interface: something players could read well enough to pay for, and platforms could recognize well enough to license.
Platforms licensing the neural-net engine. Primary revenue channel.
Players using the app I designed. Three tiers: $0 / $59 / $125.
The engine sold better than the app around it. Ten platforms licensed the API against 120 subscribers on the product I spent ten months designing. Knowing that, I would put the design effort where the revenue was and treat the consumer app as the demo for it.
Nothing in the product let players teach each other. PLO study already happens inside stables and the groups around them, which is where my own research came from, and the product never gave those groups anything to work with. A trainer a stable lead could set drills in would have grown through the people already doing that work by hand.