Just to prove my point on AI assisted product development...
As a little weekend project I created https://companychat.uk/ from scratch in net 2 days. That is from idea to a fully functional MVP (not just a demo proof-of-concept), delivering value to the end user. Feel free to give it a go, you get 5 free tokens to play with!
It's an AI aided advanced search tool for the publicly available Companies House dataset. It can help finding leads for all sorts of campaigns and help research by answering questions and prompts such as "How many companies in Brighton were founded by under 30s last year?" or "Give me a list of companies in Bournemouth that will turn 10 years old next month!".
It's hosted on a tiny EC2 instance (£6 per month), connects to an on-premise database server via Tailscale (PostgreSQL with GIS support, hosting Companies House dataset through an ETL pipeline) and to a local SQLite database for system data.
Application stack is FastAPI with Jinja templates for the backend and HTMX + TailwindCSS for frontend rendering.
AI stack: Cursor + Claude Sonnet 3.5 for development, Langchain and Langgraph with gpt-4o-mini for AI assistance.
End to end implemented by a single engineer using Cursor with Claude Sonnet 3.5.
I think how the situation turned out and how you dealt with it, gave you a much more important life experience! Dealing with BS project blockers by clients/execs and finding a way to turn it into something positive is not something they teach at school, well done team!
One idea I keep thinking about (which I'm sure interferes with some sort of copyright law) is:
1. User buys a physical copy of a movie/music
2. Instead of sending the disc to their house, it's sent to a storage facility
3. In the storage facility a digital backup is made and uploaded to a cloud storage
4. User is given access to the backup which they can watch/listen to
5. User would be the owner of said physical copy and when they sell it, all digital backups would be erased.
Chargeback is exactly what I'm going for, but need to wait until the payment settles (couple of days) and then I can start the dispute process (called my credit card provider right after I realised UberEats is not willing to refund), which with my credit card provider is the following:
- Hold the phone for 40 minutes
- Explain the situation
- Wait for a paper letter on the post that includes an empty printed form with a return envelope
- Fill out the form with the exact information I provided on the phone
- Post the return envelope
- Wait
Fortunately I don't mind playing along and I'm pretty sure having this deliberately crazy process (instead of an online form) is a much bigger pain for my credit card provider (who are hoping for churn and people abandoning the process) than myself. Well, not me. It's not the £4, it's the principal.
Love the simplicity of the app, used it for years! However when I started working with docker-compose stacks exposing PostgreSQL port I had to uninstall it because it all got confusing.
It's an AI aided advanced search tool for the publicly available Companies House dataset. It can help finding leads for all sorts of campaigns and help research by answering questions and prompts such as "How many companies in Brighton were founded by under 30s last year?" or "Give me a list of companies in Bournemouth that will turn 10 years old next month!".
It's hosted on a tiny EC2 instance (£6 per month), connects to an on-premise database server via Tailscale (PostgreSQL with GIS support, hosting Companies House dataset through an ETL pipeline) and to a local SQLite database for system data.
Application stack is FastAPI with Jinja templates for the backend and HTMX + TailwindCSS for frontend rendering.
AI stack: Cursor + Claude Sonnet 3.5 for development, Langchain and Langgraph with gpt-4o-mini for AI assistance.
End to end implemented by a single engineer using Cursor with Claude Sonnet 3.5.