Pet · AI product · 0 → 1 · 2026
Yasnost (Clarity)
A product cycle from RAT to deployment: a PRD, decisions justified through UI laws, a working Next.js prototype.
TL;DR
From risky assumptions to a deployed AI product
- 01The problem isn't the UI: people in crisis don't need another journal, they need structure. A quiz surfaces context; an AI system (psychologist + crisis manager) generates a personal plan
- 02Decisions are justified through UI-Laws.md: Fitts's Law (CTA size), Hick's Law (3 options per step), proximity (checklist grouping) — arguments, not taste
- 03Stack: Next.js + TypeScript; the product was carried to a working prototype: quiz → dashboard → PDF → payment mocks
ROLE
TIMELINE
PLATFORM
PLATFORM
The problem
In an acute crisis — a divorce, a breakup — a person lands in informational chaos. Everyone advises something, but nobody helps structure your specific situation. The products around this space are either generic articles or full therapy.
Hypothesis: there's a niche need for "structure right now, specifically for me" — somewhere between a Google search and a psychologist.
Methodology: RAT before code
Before writing any code — a RAT (Riskiest Assumption Test): identify the single assumption that, if false, kills the product outright.
The riskiest assumption: a person in emotional crisis is willing to take a quiz and trust an AI to structure their situation. If not, no stack would help.
The RAT result → PRD (2026-04-18): confirmed through desk research (comparable products, forums, behavior at crisis moments), with hypotheses formed for a user test.
Product decisions justified through UI laws
In parallel with the PRD, I wrote UI-Laws.md — a document justifying key interface decisions through principles: Fitts's Law (the main CTA is large and in thumb reach), Hick's Law (no more than 3 options per quiz step, or the user drops off), the Law of Proximity (the checklist groups by spacing alone, no dividers).
The document's goal: decisions are arguments, not designer taste. Easy to challenge, easy to revisit as data changes.
What's inside
The quiz onboarding adapts to answers: different branches for "breakup" vs. "divorce", different dashboard tone. The AI system simulates two experts: a psychologist offers a safe scenario, a crisis manager gives a radical 7-day plan plus a checklist.
The output: a clean, mobile-adaptive dashboard → download a PDF → payment (mocked in the prototype; a YuKassa integration is ready).
Stack: Next.js + TypeScript. The architecture is built for scale: new crisis types can be added without rewriting the logic.
What I learned
The RAT methodology forces an uncomfortable question before you start: "what has to be true for this to work at all?" Without it, it's easy to fall in love with a solution and build something nobody needs.
UI laws as a living document are more useful than keeping the reasoning in your head. When you need to defend a decision to a team, or return to it a month later, the document beats memory.