AI Solutions
Language models on your own servers.
We deploy and tune language models on your premises or national cloud. Nothing is sent to an outside AI service.
Where we have done it: For government agencies and cybersecurity companies. Three delivered, more being built now.
Overview
Government and security clients set one condition before they use AI: their data cannot leave their control. Public AI services do not meet that condition.
We build private LLM systems for government agencies and cybersecurity companies. Three are delivered and more are being built now. In each one the model, the document index and the applications run on client infrastructure, with no connection to outside AI providers. Client names are under NDA. References are available to qualified buyers.
In practice
Where everything runs
What we deliver
Scope of work.
- Model selection and sizingChoice of open model and GPU sizing for your users and document volume.
- On-premise or national cloudInstalled on your servers or a national data centre.
- RAG integrationConnected to your documents, with sources on every answer.
- Fine-tuning where it helpsTuning on your terminology and formats when retrieval alone is not enough.
- Security and auditRole-based access, single sign-on and a log of every query and answer.
- Operation and updatesMonitoring, model updates and support under SLA.
How we deliver
Our process.
Questions
Common questions.
Does any data leave our environment?
No. External AI services are not used.
What hardware is needed?
It depends on users and documents. We size it during assessment and can use existing servers or national cloud.
Is a private model as good as the public ones?
For answering from your own documents, a well-sized open model with good retrieval works well. We test it on your data before you commit.
Who owns the system?
You do. Code, configuration and documentation are handed over.
More in AI Solutions
Talk to us about your project.
We reply within 4 hours. Priority requests within 1 hour. Meetings on Zoom, Microsoft Teams or Google Meet.