Services
AI where it removes real work, not where it demos well
AI and automation
AI put to work inside the processes you already run: documents read and filed automatically, repetitive steps removed, internal assistants that answer on your own data, MCP servers that let AI act on the systems you use every day. Not a chatbot bolted onto the website: software that takes work off people's hands.
Most AI projects fail for the same reason: they start from the technology instead of the work. Someone buys a license, runs a pilot, and six months later nothing has changed, because nobody mapped where the hours were actually going. The hours are almost never in the glamorous part. They're in the invoice retyped into the ERP, the order confirmation copied by hand, the quote nobody can find, the report rebuilt every Monday morning from four spreadsheets.
That's where I start. Over a video call I follow the process as it really happens, and come back with a short list of the steps where a model plus some plain software can take over, and the ones where it can't, which matter just as much to know. Then I build it, connect it to the systems you already have, and stay until it's running in production with the people who use it every day.
What I build
Document and data extraction
Invoices, delivery notes, orders, contracts, technical sheets: read, understood, turned into structured data and pushed into your ERP, your management software or your database. With validation rules, flagging of what the model isn't confident about, and a human checkpoint where the cost of a mistake is high. The goal is not to remove the person, it's to stop the person retyping the same fields every day.
Process automation
The sequence of steps that today lives in somebody's head and in a chain of copy-pastes: order arrives, data gets checked, document gets generated, systems get updated, the right person gets notified. Built as real software with logs, retries and alerts when something goes wrong, not as a fragile chain of no-code blocks that breaks every couple of months and nobody can debug.
Internal assistants on your own data
An assistant that answers on your documents, your procedures, your historical data, and says "I don't know" instead of making something up. Access controlled by role, answers that cite the document they came from, and your data staying where you decide it stays. Useful when the knowledge of how things work is locked in the heads of two people and everyone else keeps interrupting them. Whoever the assistant talks to, the notice that they are talking to a machine is there from the first reply: the EU AI Act has required it since 2 August 2026 and it ships as standard, not as a paid extra.
MCP servers and AI connected to your systems
An MCP server is what lets an AI model actually act on your software: read from the ERP, create a ticket, update an order, pull a report. I build the server, the permissions and the audit trail, so the AI can do the work inside your perimeter instead of being handed an API key and hoped for. The same infrastructure that makes an AI agent useful is what keeps it from doing damage.
The measure of this work is not how advanced the model is. It's how many hours a week your people get back, and whether the process still runs correctly on the Monday I'm not there.