Custom AI Applications

A custom AI application starts at $8,900, quoted after a scoping workshop. The price is driven by how complex the logic is, how many of your systems it integrates with, and what your data handling and GDPR requirements are. AI model usage is billed separately, so running costs stay transparent and under your control.

What I have actually shipped

This is not a theoretical offering. Four live AI applications, used by tens of thousands of people:

• PixiFixi — AI photo editor, 30,000+ users since September 2025
• Ogrovision — the first AI garden visualiser in Poland, 20,000+ users
• Lovelio — the first Polish AI greetings generator, 10,000+ users in a year
• Izbaviz — the first AI interior visualiser in Poland

All four went from idea through launch to running under real traffic. That is a meaningful difference from a contractor who delivers and disappears before anyone finds out how the thing behaves with a thousand users a day.

What building them taught me

The model is the easy part. Everyone has access to good models now, through an API, for pennies to start. The difficulty is everything around it: what to show when the model returns nonsense, how to keep cost per user viable, what happens when your provider raises prices or retires a model version.

Interface decides adoption. I design for people who do not know what a prompt is — and do not need to. Upload a photo, pick a preset, get a result. That audience is consistently overlooked by AI tool makers, and takes to these tools readily once they stop being intimidating.

Model costs have to be planned before the first line of code. An app costing pennies at a hundred users can eat the entire margin at ten thousand. I bill model usage separately from the build precisely so you see that cost directly and can control it.

When a custom application is worth it

Worth it when no existing tool does what you need and the process repeats often enough for the saving to compound. Or when you want a product rather than an internal improvement — as with the four above.

Not worth it when an off-the-shelf tool on a subscription already does eighty per cent of the job. A custom application is the build cost plus maintenance forever. I do talk people out of building and point them at existing tools instead — I would rather do that than take on a project that cannot pay for itself.

Also not worth it when you are not yet sure what the application should do. It is cheaper to start with simple automation on existing tools, test the assumptions in practice, and build afterwards.

How a build runs

It starts with a scoping workshop where we settle the scope and check the idea makes economic sense. Only then do I quote — anything earlier would be guesswork.

After that I work in stages, beginning with the narrowest version that functions. Not a full feature set, but the one flow that proves the whole thing holds together. Expansion happens on working foundations rather than on assumptions.

Data and GDPR

With AI applications this is not paperwork. We settle up front whether your data can reach an external model provider, whether a processing agreement is needed, and whether a self-hosted model is required instead. Being EU-based, I treat this as a design input rather than a compliance step bolted on at the end — and it genuinely changes both architecture and cost.