Prompt-to-Design Commerce MVP
Users generate wearable designs from prompts with live preview and saved sessions.
Overview
A solo build exploring AI-assisted design → purchase flows with a friction-light canvas UI.
Context & Role
Solo PM/IC: market hypothesis, UX, MVP scope, validation loops.
Problem
Cold-start creativity; session persistence; safe prompts.
Objectives
Fast ideation → preview; save/share; low latency.
Product Decisions
Prioritised latency and usability over advanced generation controls. Kept profiles RLS-secured for safety and portability.
Solution
React + Canvas customiser; GPT for ideation; Supabase Auth/RLS; order model (size/SKU/pricing); token accounting + retries.
Architecture
React FE ↔ Supabase (Auth/DB/Storage) ↔ Prompt service ↔ Order service.
Metrics & Impact
100+ testers; 3.4 designs per first session; qualitative feedback: playful & fast.
Evidence
Event logs (session length, saves); prompt error rates; order funnel.
Challenges
Prompt safety; rendering performance on low-end devices.
Lessons
Preview latency trumps model fanciness for novice users.