• Taking ownership of the model lifecycle - pipelines, registry, versioning, CI/CD, drift detection, and retraining logic
• Working out how to run models in constrained or edge settings: quantisation, pruning, distillation, ONNX/TensorRT, and picking the right accelerators for the job
• Setting the evaluation approach - where precision and recall trade off, where thresholds sit, and how much false-positive friction is acceptable for the people using the system
• Designing so the model supports human decision-making rather than overrides it
• Building for sites with poor or intermittent connectivity, with training centralised and models pushed out from there
• Getting the non-functional side right at scale - throughput, latency, availability, resilience, DR - across multiple sites
• Reviewing and challenging supplier architecture, including calling out vendor lock-in risk and build-vs-buy calls
• Working to Secure by Design principles and NCSC guidance throughout, given the OFFICIAL-SENSITIVE classification

What you'll bring

• A track record as a solution/technical architect on AI projects sitting inside bigger strategic programmes
• Solid AWS experience - S3, Lambda, EventBridge, SNS/SQS
• Comfortable working across TOGAF, ArchiMate, C4, UML and BPMN
• Hands-on with MLOps tooling and platforms - SageMaker, OpenVINO or equivalent
• Experience architecting data for large volumes of imagery - tiering, retention, lineage, provenance
• Familiar with the governance side of AI - DPIAs, model documentation, bias and fairness checks
• Strong stakeholder handling, and the ability to explain architecture to people who aren't architects