Before your organisation commits a single euro to hardware, we build your use case on our lab with your real documentation. You see the result, measure whether it serves you, and decide on your own data.
Why this service exists
The question that blocks almost every local AI project is not technical, it is one of trust: will this actually work with my material? Nobody can answer that with a generic demo over sample documents, and nobody should have to buy a server to find out.
We work on our own NVIDIA DGX Spark with open models. It is the right machine for prototyping and measuring; production sizing is a separate conversation, and we have it when the report is delivered.
What you get
A working demo over your corpus, reachable so that the people who would actually use it can try it, not just the technical department.
A measured quality report. A set of questions representative of your operation, with the answers obtained and an accuracy assessment, not a general impression.
Production sizing. What hardware you would need for the concurrent user count you gave us, with alternatives and their rough fit.
Corpus diagnosis. Which part of your documentation is ready, which needs preparation and how much work that means.
An honest recommendation, including not going ahead if the numbers do not add up.
How we handle your documentation
This is what we get asked about most, and rightly so. We are a cybersecurity company: if we did not take care of this, we would not have much to offer.
1
NDA first
Before a single file reaches us. With scope, retention period and a written destruction commitment.
2
Isolated environment
Your corpus is processed in a dedicated environment, with no internet access during processing and separate from any other client.
3
No third-party egress
Models run locally. Your documentation is not sent to any external service and does not train anything.
4
Verifiable destruction
When the test closes, corpus, indexes and copies are wiped, and we certify it in writing.
If you would rather not hand over real material, we work with anonymised documentation or a bounded subset. The measurement loses some fidelity, but it remains far more informative than a generic demo.
What we need from you
A concrete use case. “Answering things” is not a use case; “letting the support team find the procedure that applies to an incident” is.
A representative sample of the documentation that would feed the system.
Between 20 and 30 real questions your people ask daily, with the correct answer known. That is what lets us measure instead of opine.
An estimate of concurrent users so production can be sized.
What it is not
This is not a production pilot and must not be used for real decisions. It is a test bench to measure feasibility and quality. No high availability, no directory integration and none of the controls a final deployment carries.
Frequently asked questions
How long does it take?
Two to three weeks from receiving the corpus and the control questions, depending on volume and the state of the documentation. Most of the time goes into preparing the corpus, not into standing up the model.
What if the result is poor?
Then the test did exactly its job, for a fraction of what finding out with the hardware already bought would have cost. The report will explain whether the problem is the use case, the corpus or the sizing, and whether it is fixable.
Can we see the system running before commissioning the test?
Yes. We show you the lab with public documentation in a half-hour session. Not enough to decide on, but enough to understand what we are talking about.
Does the test carry over if we go ahead?
Yes. Corpus preparation, model selection and the control question set are reused in full during deployment, and discounted from the scope.
Try it on your own material before buying anything
Tell us the use case you have in mind and we will say whether it is a good candidate, what we would need and how long it would take. No commitment and no forms.
Phone: 686 250 244 · Email: info@jaymonsecurity.com
You may also be interested in
