Building a smart system
you can actually work with.

What can the system include?

• AI connected to documents and data sources
• Agents that carry out defined processes
• Users, permissions and an admin area
• History and data storage
• Payments, subscriptions and integrations

What do you get?

A spec for the first version, infrastructure planning, system development, testing, deployment and an orderly handover of access and code.

Who is it for?

For a business that wants to turn knowledge, a workflow or an idea into a complete AI system — not just add a small chat to an existing site.

How is it priced?

A system like this is priced after a discovery call and defining the first version. The price depends on the data, the number of users, the level of permissions and the external integrations.

Discovery call

We start with a 30 to 45 minute call. I ask which process you are trying to solve, who the users are, where your data lives today and what happens when something breaks. The call produces a first spec for version one — not for the perfect system.

Spec and infrastructure planning

We define what goes into version one and what waits. We choose the database, the model, the retrieval approach and the permission model. By the end of this step it is clear what the system does, how long it takes and what it costs.

Development in versions

We build in stages and deploy to a test environment early. You see real screens within weeks, not at the end of the project. Every change of direction costs less the earlier it arrives.

Deployment and handover

We go live, run basic load and security checks, and hand over access, code and documentation. If you want to keep developing on your own, everything is already yours.

How long does it take to build an AI system?

A working first version usually takes 4 to 10 weeks. The range depends on the number of data sources, the number of user types and the integrations. A system with one data source and one user type lands at the low end.

What is the difference between RAG and uploading documents to a chat?

RAG indexes your content and retrieves only the passages relevant to each question. You get answers that cite a source, it works across thousands of documents, and it does not mix data between customers. Uploading files to a chat is fine for personal use, not for a system.

Our data is sensitive. Where is it stored?

Data stays in infrastructure you control — usually Supabase or a cloud provider you already use. We can make sure the model does not train on your data and limit which passages leave your environment at all. This is decided during the spec, not afterwards.

What happens after handover?

You can stop there — the code and the access are yours. If you want, I am available for monthly support covering fixes, improvements and additions. There is no required retainer.