Prompt Bnk

LiveJune 2025 to March 2026

A UK startup wanted people with little experience of writing prompts to get better, more reliable answers from language models. I built Prompt Bnk, a subscription library where they find a prompt, copy it and use it, and I curated the 150 prompts in it.

Searching the Prompt Bnk library for plan, clearing it, then filtering by one category

What I did

Architecture and backend
I designed the architecture and built the whole API, including subscriptions, access control and the admin tools.
All of it
Interface
I designed and built the interface, with discovery by category and access gated by subscription.
All of it
The prompts
I curated the 150 prompts across 15 categories and rated each on five measures: clarity, consistency, relevance, efficiency and effectiveness.
All of it
Testing and release
I tested it and took it live.
All of it

How it works

Browse
The prompts sit in 15 categories, so you start from the kind of work in front of you.
Trust it
Every prompt was rated on clarity, consistency, relevance, efficiency and effectiveness before it went in.
Subscribe
It is a subscription library. Signing up opens the prompts.
Copy and use
Copy a prompt and paste it into the AI tool you already use.
Stay current
When the library changes, the page you have open updates by itself.

In the browser

01Interface
Prompt discovery by category, and access flows gated by subscription.
- React 18
- TypeScript
- Tailwind CSS
- Lucide React

Requests in, live updates back over SignalR

The API

02Backend
A modular monolith, with payments and analytics as modules. The library is held in PostgreSQL through Entity Framework Core.
- ASP.NET Core
- .NET 9
- Entity Framework Core
- PostgreSQL
03Subscriptions and access
Subscription management and access control decide what an account can open.
04Live updates
SignalR pushes library changes to every open page, so a new or edited prompt appears without a reload.
- SignalR
05Logging and metrics
Structured logs through Serilog. Usage metrics track how each prompt is used and how well it works.
- Serilog

Simplified. The admin tools, sign-in and the delivery pipeline are left out.

What testing found

In testing, people completed 19 of 20 tasks. The median time to the first useful prompt was 32 seconds, and the copy rate was 90%.

They scored its usability 84.5 out of 100, its usefulness 86 and its ease of use 88.

Over 50 people subscribed in the launch period.

19 of 20
Test tasks completed
32s
Median time to the first useful prompt
90%
Copy rate in testing
84.5
Perceived usability, out of 100
86
Usefulness
88
Ease of use
50+
Subscribers in the launch period
150
Prompts I curated, across 15 categories

19 of 20 task attempts succeeded, with a median of 32 seconds to the first useful prompt. Usability averaged 84.5 out of 100. Usefulness and ease of use were rated separately, at 86 and 88.

Usage metrics in the backend track how each prompt is used and how well it works, alongside the five measures each prompt was rated on when I curated it.