Senior AI Systems Engineer

We regulate financial services firms in the UK, to keep financial markets fair, thriving and effective. By joining us, you’ll play a key part in protecting consumers, driving economic growth, and shaping the future of UK finance services. 

The Data, Technology and Innovation (DTI) division enables the FCA to be a digital-first, data-led smart regulator by delivering a secure, agile, and cost-effective technology and data ecosystem that drives better decisions, transparency, and operational efficiency.

The AI Product Delivery Department (AIPD) exists to transform FCA business processes through AI, safely and at scale. It provides the product-led capability to meet growing organisational demand, moving work from promising AI experiments into live, adopted services. Its operating model combines portfolio and value management, platforms and engineering, and three cross-functional delivery pods focused on Authorisations, Supervision and Enforcement priorities. The Products and Delivery Team owns pace, execution and adoption: turning validated business problems into secure products, embedding AI into real workflows, using evaluation, provenance and go-live discipline

You will build and operate the intelligence and orchestration layers of FCA AI products. The role is software-engineering led, not prompt-development led: it makes model, retrieval, tool and orchestration behaviour testable, evidence-grounded, observable and reliable in production.

• →Build and operate production-grade AI systems using foundation models, retrieval, tool-use and workflow services that solve real regulatory challenges and are embedded directly into operational FCA workflows, taking ownership of model and tool integration, retrieval and context assembly, provenance mechanisms, workflow orchestration, evaluation instrumentation and failure handling

• →Develop advanced retrieval-augmented generation capabilities including context and retrieval pipelines, semantic search, structured outputs, evidence references and citation verification to deliver evidence-grounded and auditable outcomes that supporting sources can be inspected where required by the relevant risk tier

• →Design multi-step and agentic workflows with human review, approval, escalation and safe-failure mechanisms, helping shape how AI is applied safely and responsibly within a major financial regulator while implementing explicit controls for known failure modes and appropriate operational monitoring

• →Build agentic orchestration covering planning and replanning, structured tool use, persistent state, semantic and episodic memory, agent hand-offs, execution limits, human interruption and recovery from failed or incomplete tasks

• →Partner with stakeholders to create evaluation datasets and automated testing frameworks that measure task success, retrieval quality, source support, robustness, latency, cost efficiency and regression performance

• →Champion secure engineering practices by building resilience against prompt injection, unsafe tool use, access-control failure, data leakage, unsupported claims, retrieval failures and model or configuration degradation using appropriate third-party tooling and governance controls

• →Collaborate with Platforms and Engineering team, architecture, product and domain specialists across Authorisations, Supervision, Enforcement and AML, contributing to CI/CD, release management, runtime operations, engineering quality, supplier oversight, knowledge transfer and permanent organisational capability building

• Solid Python and production software engineering, including APIs, automated testing and cloud-native services
• Practical experience building and operating production applications using foundation models and LLM APIs
• Experience with retrieval and embedding pipelines, semantic search or equivalent AI data pipelines

• Evaluation and monitoring of AI systems as a disciplined practice, including golden datasets, regression suites, retrieval assessment and citation quality measurement

• Designing evidence-grounded, human-authorised workflows with explicit failure handling, abstention criteria and escalation pathways

• Applying working knowledge of AI failure modes, including hallucination, retrieval drift, context failure, prompt injection and performance degradation, together with appropriate controls

• Implementing CI/CD, version control and operational telemetry with prompts, configurations, datasets and evaluation assets managed as versioned production artefacts

• Delivering secure engineering practices, including access control, data leakage prevention and safe tool usage

• Communicating technical trade-offs within multidisciplinary teams, reviewing supplier deliverables and ensuring effective knowledge transfer

• Particularly valuable: Experience with AWS Bedrock and AgentCore or comparable cloud and foundation model platforms, alongside agent orchestration, event-driven systems, graph-enhanced retrieval and temporal and provenance models. Practical experience in evaluating complete agent trajectories and memory behaviour, including task completion, tool selection, escalation, memory relevance, continuous improvement from prior outcomes, adversarial testing, prompt-injection red teaming, policy-aware retrieval. Delivery within regulatory technology, financial services or other controlled operational environments

• 25 days annual leave plus bank holidays
• Colleagues spend a minimum of 50% of their working time in the office each month (60% for Directors and Executive Directors) across our London, Leeds and Edinburgh offices
• Non-contributory pension (8–12% depending on age) and life assurance at eight times your salary

• Private healthcare with Bupa, income protection, and 24/7 Employee Assistance

• 35 hours of paid volunteering annually

• A flexible benefits scheme designed around your lifestyle

For a full list of our benefits, and our recruitment process as a whole visit our benefits page.

Our colleagues are the key to our success as a regulator. We are committed to fostering a diverse and inclusive culture: one that’s free from discrimination and bias, celebrates difference, and supports colleagues to deliver at their best. We believe that our differences and similarities enable us to be a better organisation – one that makes better decisions, drives innovation, and delivers better regulation.

If you require any adjustments due to a disability or condition, your recruiter is here to help - reach out for tailored support.

We welcome diverse working styles and aim to find flexible solutions that suit both the role and individual needs, including options like part-time and job sharing where applicable.

We’re proud to be a Disability Confident Employer, and therefore, people or individuals with disabilities and long-term conditions who best meet the minimum criteria for a role will go through to the next stage of the recruitment process. In cases of high application volumes, we may progress applicants whose experience most closely matches the role’s key requirements.

• Advert Closing: 11 October at 11:59pm

• CV Review/Shortlist:  13 October

• Technical Assessment: 14 - 21 October

• Case Study: 29 October- 3 November

• Interview W/C: 9 November

• The selection process will involve a technical assessment, a case study presentation and a formal competency-based interview

• Your Recruiter will discuss the process in detail with you during screening for the role, therefore, please make them aware if you are going to be unavailable for any date during this time.