Auth and access via Supabase
Supabase handles email + password sign-in, plus row-level security so the library can scale to other accounts later without rewiring the data layer.
A prompt library that turned AI from a daily experiment into a daily habit
Years
Q1 2026 — ongoing
alongside Memorisely AI Design + AI Coding course
Stack
Figma Make · Supabase (auth + Postgres)
My role
Solo — design, build, content
In one line
AI stopped being a side experiment and became part of my UX week.
A prompt that worked once is not the same thing as a prompt I can rely on
The pattern was simple. I'd write a good prompt in ChatGPT for synthesis or copy review, get a useful result, forget the exact wording, and a week later end up with a version that almost — but not quite — worked. The fix wasn't smarter prompts. It was treating them like any other reusable asset — named, versioned, with a note on what they're for.
“If I'm rewriting the same prompt every week, it's not a prompt — it's an asset I haven't named yet.”
Five categories — the parts of UX where AI actually helps
- Working with research notes. Turning interview transcripts into themed maps I can walk into a 30-minute review with.
- Drafting and reviewing copy. First-pass labels, tooltip and validation copy. A human edits every line before it ships.
- Pressure-testing layouts and labels. Asking what a stranger with no context would think a label means.
- Pressure-testing my own thinking. Before scheduling the real review, asking what the strongest objection to this design would be.
- Onboarding myself into a new domain. Closing the distance between “I don't know this domain” and “I can ask the right questions.”
The categories aren't aspirational. They're where I'd already started leaning on AI before the library existed. The library made each one repeatable.
Figma Make on the front, Supabase on the back, zero servers in the middle
Versioned prompts in Postgres
Each prompt carries its category, a working note (“what this is for”) and a version stamp. When a model update changes how a prompt behaves, I can see exactly which version started failing — and why.
A read-only demo account — see the library without a screen-share
The credentials below open a demo account scoped to the prompt library content. They can see the shape of it but can't change anything. The site lives on a Figma Make domain — this is a working tool, not a polished marketing surface.
nikitakesha42@gmail.comshokoladnyjmuss96069A small tool with three quiet effects
Product
5 categories
Versioned prompts I reach for daily. No servers to maintain. Honest scope — this is a personal tool, and that's the point.
Process
Prompt-as-asset
Treating prompts like reusable assets — named, versioned, intent-noted — changed how I write them in the first place. The “rewrite from scratch” tax dropped to near zero.
People
A working pattern
When teammates ask “how are you actually using AI”, I can show them the library instead of waving at it.