Fixed scope, duration and price
We set them with you before we start. The price depends on the case, so we quote it on request.
Collaborate
We start with a pilot, on your server
One closed format to test one of our lines with your data and on your infrastructure. If it works, we continue it as an R&D project.
Tell us about your caseThe on-premise pilot
An agreed metric
Before we start, we agree which number the pilot has to move to count as a success.
Your data, your server
We work with your data and on your infrastructure, with our own stack.
What you get
A working system, a report and the decision to continue or stop.
After the pilot
If the pilot works and you want to go further, we continue it as a phased R&D project and, if it fits a funding call, we look for funding for it. We never start there.
Which pilots we take on
- Only from our three lines (on-premise AI, data and agents), built on our stack and leaving something reusable.
- We open at least two pilots per quarter; how many more depends on the size of each one.
- No pilot with a clinical purpose until we have the CE marking.
What we work with
Only what we have tested and written about in the Sprint DGX series.
NVIDIA DGX Spark
Why the model lives in the clinic, not in the cloud.
Read the chapter →vLLM
From the trained model to one that starts answering in milliseconds.
Read the chapter →Gemma 4
Why we chose it as the base model over LLaMA 3.2 Vision.
Read the chapter →VLM fine-tuning
How we teach dentistry to a vision-language model, stage by stage.
Read the chapter →Tell us about your case
What you want to solve and which data cannot leave your organization. We reply within 24-48 h.