Sovereign and private AI deployment
Capable AI on infrastructure you control, run by your own team.
Open-weight models served on-premises, in an in-country cloud, or air-gapped, evaluated on your data and handed over. For most serious work in Saudi Arabia, data residency is a condition of the engagement, and this practice is built around it.
The data stays inside your boundary. So does the model.
02What it is
The ArchiTechs Group deploys open-weight models from the Llama, Qwen, Kimi and DeepSeek families on infrastructure inside Saudi Arabia that the client controls: on-premises, in an in-country cloud, or air-gapped. Each deployment is evaluated on the client’s own tasks, sized to the workload, monitored, and handed to the client’s team. Private, in-Kingdom deployment is the default requirement for much Saudi work, and building that capability locally is part of what Vision 2030 asks of the sector.
For whom
For any organisation whose data cannot leave the Kingdom or its own network: government, healthcare, finance, legal, and critical infrastructure.
03Shape of the work
Candidate models benchmarked on your own tasks. If a hosted API is the better choice for non-sensitive workloads, the evaluation will recommend it and the engagement can end there.
Serving stack, hardware sizing and network boundary, drawn so your security team can review it before procurement.
Your documents indexed inside your boundary, so answers are grounded in your own material and nothing leaves the network.
Policy, rate limits, and an audit log with the retention period your compliance function requires.
Runbooks, an upgrade path, and your own engineers operating the system.
- 01Evaluation firstCandidate models benchmarked on your own tasks. If a hosted API is the better choice for non-sensitive workloads, the evaluation will recommend it and the engagement can end there.
- 02Deployment architectureServing stack, hardware sizing and network boundary, drawn so your security team can review it before procurement.
- 03Retrieval over documentsYour documents indexed inside your boundary, so answers are grounded in your own material and nothing leaves the network.
- 04Guardrails and audit logPolicy, rate limits, and an audit log with the retention period your compliance function requires.
- 05Hand-overRunbooks, an upgrade path, and your own engineers operating the system.
04What you receive
How the engagement runs
Fixed-scope, beginning with the evaluation, so the decision rests on evidence from your own tasks.
- Models serving on your own hardware
- The evaluation suite that selected them
- Architecture documents reviewed by your security team
- Guardrails, monitoring and the audit log compliance requires
- A team that can run it with no dependency on us
05The other lines of work
All services- 01
Enterprise systems and ERP
An ERP that fits how the organisation works, and still upgrades.
- 02
Web and mobile applications
Applications people use every day, in Arabic and in English.
- 03
Fintech and regulated platforms
Payment, lending and back-office platforms, built to be audited.
- 04
AI transformation
From assessment to systems in production, with one team advising and building.
- 05
AI governance, safety and compliance
AI you can defend to a regulator, with the controls in place.
06Start here
If your data cannot leave your network, start with an evaluation on your own tasks.
Tell us what the data is, where it must stay, and what you want a model to do with it. The first reply says what an evaluation on your own tasks would involve.
Start a conversation