CASE STUDY
SOLO PROJECT · DESIGN + BUILD · 2025–2026 · AI PRODUCT
LineShift
Sports betting has a decision problem, not an information problem.
OVERVIEW
LineShift reads live odds across four sportsbooks, finds where a price is soft, and hands you a plan you can actually act on: which book, what stake, what to watch. I designed and built it end to end, product thinking through front end.
THE PROBLEM
01 — The information already exists. Nobody can act on it.
Odds, injuries, weather, line movement are all public. They are spread across a dozen tabs and they move every few minutes. By the time you have assembled a picture, the price is gone.
02 — Every tool in this space overclaims.
Betting products sell certainty because certainty converts. Locks, guaranteed plays, hit rates with no sample size behind them. It works right up until a user loses money and realises the confidence was decoration.
03 — Knowing what to bet still leaves the hard part.
Even once you’ve found a good number, you still have to decide how much to put on it, which book to use, and whether it’s still worth it an hour later when the line has moved. Most tools stop at the pick.
HOW I DESIGNED IT
A plan, not a pick.
Every recommendation resolves to a structured ticket: the market, which of the four books has the best line, a stake sized off bankroll rather than gut, correlation between legs, and what to monitor before it starts.
The AI Bet Builder
Confidence had to be legible, not loud.
The decision I spent most time on. Each candidate shows signal strength as a fraction, 6 of 9, instead of a percentage. A fraction tells you how much evidence exists. A percentage implies precision the model doesn’t have.
Six of nine checks cleared, and which three didn't
New accounts start empty.
There is nothing to show someone on day one, so instead of blank panels every screen explains what will appear there and what you need to do first. It ends up doing the job of onboarding.
A lot of information on one screen, on purpose.
There are a lot of numbers here and I didn’t try to hide them. What matters is the order: the strongest play at the top, supporting markets below it, the book comparison to the side. If you’d need something while you’re deciding, it’s visible without clicking.
WHAT I DECIDED AGAINST
Every product in this category ships a lock of the day.
The obvious version of this picks one bet, calls it the play, and puts a big confident number next to it. That’s what many betting tools do, and they convert, because people want to be told what to do.
I built the opposite. Every pick runs through nine checks: line movement, injury impact, weather where it matters, disagreement between books, and so on. The card shows how many it cleared (ex: 6 of 9), and which ones it didn’t.
The cost is real. 6 of 9 is less satisfying than LOCK OF THE DAY. But there is no such thing as a guaranteed lock, and I wanted to prioritize transparency and delivering a tool bettors can trust.
Most nights, most picks score low. The board says so.
WHAT I’D DO NEXT
Building up a real track record
The Performance page doesn’t show a win rate until there are 30 graded picks, which means the product can’t prove anything about itself until it’s been used for a while. Getting that history is the next real milestone. After that I want it to explain in plain language why a price moved rather than just showing that it did, and to tighten the mobile flows, since the bet actually gets placed on a phone.