OneAdvanced Builds UK-Sovereign AI Platform with Self-Hosted Llama Models on AWS

OneAdvanced, a UK-based enterprise software provider serving over 10,000 customers across regulated industries, has built a UK-sovereign AI platform by self-hosting open-weight large language models (LLMs) on AWS infrastructure they control. This approach was driven by the need to ensure data residency and compliance, as their customers handle sensitive information such as health records and legal documents.

Initially, the company prototyped with Amazon Bedrock, but to meet sovereignty requirements, they opted to deploy models like Llama 4 Maverick and Llama Guard 4 directly on AWS in the UK region, as these models were not yet available through managed services. This involved deploying, serving, and scaling the models on Amazon SageMaker AI, with infrastructure running on p5.48xlarge instances.

The architecture includes a Retrieval Augmented Generation (RAG) pipeline backed by Amazon Aurora PostgreSQL with the pgvector extension, enabling document retrieval and embedding. Over 50 specialized agents, built with the Strands Agents SDK, run on Amazon ECS, each with custom prompts and tool configurations, supporting diverse use cases across sectors.

Security and compliance are integral to the platform, with data stored exclusively within the UK, no user data retained or used for training, and content moderation handled by Llama Guard 4. The infrastructure is monitored with Amazon GuardDuty, and the company holds ISO 42001 certification for AI governance.

Since deployment in July 2025, the platform has been operational, serving customers with high satisfaction scores. It demonstrates how self-hosted models can meet strict data sovereignty standards while delivering practical AI capabilities for enterprise use in regulated industries.