References

Four things that are built and running

Client projectClient not named

Supplier price lists from email straight into the database

The brief
Hundreds of rows a day from supplier price lists, retyped by hand out of email.
The result
The price lists process themselves and the data is consistent, with no typos. Verified on the client’s real price lists.
My own product, live since June 2026In service

Website and admin for the bikested app

Evidence
Publicly at bikested.com
Sector
App for cyclists
The brief
An app for cyclists in the App Store with no website and no admin where I could see what happens inside it.
The result
A website in two languages with links to the App Store and Google Play, and an admin area with logins where I follow how the app runs and what the AI costs.
My own product, in the App Store since June 2026In service

AI mechanic in the bikested app

Evidence
Available on the App Store and Google Play
Sector
App for cyclists
The brief
A cyclist asks about their own bike: when to replace the chain, whether a part fits, whether a damaged part is safe to ride on. A general chatbot knows neither their bikes nor the parts catalogue.
The result
The assistant knows the user’s garage and looks parts up by itself in the database and the knowledge base (RAG). The model was picked by a test on 87 questions, including photos of damaged parts.
My own internal systemIn service

My own sales system: enquiries, CRM and reminders

Evidence
Run logs in n8n and in the database
The brief
Enquiries from the website and from email, phone calls and sent emails, each in a different place. Nothing showed who was waiting for a reply.
The result
Nine n8n workflows, in service since 31 Aug 2026. AI reads an enquiry from email and logs it in the CRM, and when I do not reply in time, a reminder arrives. Replies and calls stay with me.

No client names. I do not have their permission to use them as a reference.