Open roles

Come build sovereign AI solutions

Built for healthcare, where you have to prove what the model did and where the data went. You'll extend the layers - and build the products that sit on them.

OpenAI Anthropic Google Blocked
Local AI‑infrastructure Full sovereignty Multi-LLM · 27+ open-source models GPU / CPU · GDPR & AI Act-aligned
DeepSeek Baidu ByteDance Blocked

Why we build AI this way

We own the whole thing

From the machines in the datacentre to the app on the screen, we own the layers rather than rent them. Nobody in the middle can change the price, the rules or the model underneath us.

Because it's people's data

What we handle is about as personal as data gets. If you're going to ask people to trust you with that, you should be able to point at where it lives and who can reach it.

And it makes the work better

When you own each layer you can actually fix things - swap a model, trace a request, change how something behaves. No support ticket to a vendor, no waiting for someone else's roadmap.

One stack. Eight layers we run ourselves

Eight layers, from the application at the top down to the metal - and the same identity, consent and audit trail across all of them.

  1. 01 Application Your products - clinical AI, research tools, analytics
  2. 02 Orchestration Router, load balancer, workload manager
  3. 03 Model layer 27+ open models - Llama, Mistral, Mixtral, Gemma, Phi
  4. 04 Data layer Vector DB, object storage, audit ledger
  5. 05 Infrastructure software Kubernetes, vLLM, PyTorch, TensorFlow
  6. 06 Network Private VPC, VPN, direct connect
  7. 07 Compute Our own GPUs, GPU scheduling, InfiniBand
  8. 08 Datacenter European datacentre, private network, green power

Data flows in. Answers flow out. All of it inside our perimeter.

Stack, models and audit trail live inside one perimeter we control end to end. No connections to the US or China - only your tunnel in and the response back.

USA · BLOCKED OpenAI Anthropic Google CHINA · BLOCKED DeepSeek Baidu ByteDance 100% SOVEREIGN Models 27+ LLMs Compute GPU / CPU Data RAG · vector Audit SHA-256 Your prompt from your app VPN TUNNEL Response audit-signed RETURN

How a request actually flows

You write one prompt. We orchestrate the rest.

  1. 01

    Prompt

    Your application sends one request. No model name, no vendor SDK.

  2. 02

    Router

    The routing engine reads the task type and picks the model - a 7B for classification, a 70B for long-context reasoning.

  3. 03

    Model

    27+ open models on our own GPUs. Fallback strategies if one fails.

  4. 04

    Ledger

    Every call lands in a hash-chained audit ledger. That's how consent stays provable.

What you'd actually touch

See open roles
  • Inference pipelines built and optimised on our own GPU infrastructure.
  • Open-weight models, from fine-tuning through to production.
  • Bare-metal and private VPC on European soil - WireGuard, TLS, WAF.
  • CI/CD, infrastructure-as-code and observability across every brand.