Your AI, on your hardware, within the law
We start with a 15-minute call. If it fits, we scope the work, build it on your machine and leave it running and documented.
EU AI Act: deployer-ready in two weeks
For companies that already use AI (a chat assistant, a copilot, a chatbot on the website) and want to have in order what Regulation (EU) 2024/1689 asks today of whoever uses it: the "deployer".
- An inventory of your AI uses Which AI tools your team uses, for which task and with which data. Everything else builds on it.
- An AI literacy plan and one training session What Art. 4 asks, applicable since 2 February 2025: proportionate measures so the people using AI know how it works and where it fails.
- Labelling and transparency The Art. 50 notices that apply to you, applicable since 2 August 2026: saying a chatbot is an AI and labelling generated content where required.
- Risk classification Each use, at its risk level and in writing. If one is high-risk (Annex III), we tell you and explain what that means.
- A written scope, agreed on the call What is in, what is out, who on your team takes part and on which dates. A fixed written quote after the call.
Two weeks from the signed scope, if your team can make the interviews and the training session in that window. If a use is high-risk, two weeks are not enough: we tell you on the call and propose a different scope.
How we work
Four steps. You know what happens in each one before it does.
-
A 15-minute call
You tell us what you want to solve. We tell you if we can help, or if you do not need to hire us.
-
Scope in writing
We pin the work down: what is delivered, with which data and by when. A fixed quote before we start.
-
Deployment on your hardware
We install the models on your machine, measure them on your real tasks and connect them to your systems. The data stays on your network.
-
Handover and training
Documentation, training for your team and the obligations checklist (GDPR, EU AI Act) completed. Support afterwards, if you want it.
What we measured on our machine
NVIDIA DGX Spark (GB10, 128 GB unified memory), measured 2026-10-04. Each figure links to the post with the method.
- llama3.1:8b 41.3 tokens/s 9.2 GB loaded
- qwen2.5-coder:7b 41 tokens/s 6.6 GB loaded
- gemma4:26b 60.7 tokens/s 17 GB loaded
- faster-whisper small 0.25× real time 30 min of audio in about 7.5 min
Your hardware will give other numbers. That is why we measure on your tasks before recommending a machine.
Other ways of working
If the AI Act is not your need, or you already have it in order. Each one is quoted by scope, in writing and before we start.
- Audit 1–2 days
- A diagnostic of your systems, workflows and data. Deliverable: a report with actions and a prioritised roadmap.
- Build sprint 2–4 weeks
- A local model on your machine, measured on your tasks and connected to your systems. Scope agreed before we start.
- Managed support Monthly
- Monitoring, model tuning and a quarterly review, during business hours (9–18 CET).
Let's talk for 15 minutes
No commitment. We tell you if we can help, or if you do not need to hire us.
Book a 15-minute call