Hey Product Hunt 👋 Every product eventually depends on a number nobody wrote down. What this repair costs. How long this ticket sits. What this thing is worth. We kept doing the same two bad things: hardcoding a constant and forgetting about it, or asking an LLM and getting a confident number with nothing behind it. So we built maap. You send a record as plain text and you get back a number, the spread around it, and the odds of crossing whatever threshold you actually care about. No schema to define, no model to train, no pipeline to maintain. One call. It is trained with RLCE, Reinforcement Learning for Calibrated Estimations. Plainly: the range is tuned against real answers the model never saw, instead of being asserted and hoped for. The spread is the whole point. "Probably $9,000" and "probably $9,000, with a 6% chance it goes over your limit" are not the same answer, and only one of them is something your code can act on. Easiest way to see it: on the page there is a demo where you place your own guess first, and only then does the band appear. Five real things: a house, a used car, a hospital bill, a diamond, a pound of avocados. Most people lose to it. Some don't. Free to try, no card. Would love to hear what numbers you'd point it at. Genuinely, that's the most useful feedback we can get right now.

maap: the number LLMs can't guess

LLMs guess numbers. maap gives you the odds with RLCE

LLMs guess numbers. maap gives you the odds with RLCE