AI / ML Engineer - FDT in Sheffield

Job Description

Level / Location: GCB4/5/6 — Various — co-located with the FDT Tech

Sit within the FDT Tech team and close to users (business teams, delivery pods, transformation squads) as the accountable AI Engineer for rapid prototyping and delivery of advanced AI solutions. Design, build and iterate production-viable AI prototypes and thin-slice solutions spanning advanced modelling, GenAI/RAG/agentic workflows, evaluation harnesses and safety controls—turning real user needs into working AI capabilities quickly, safely and repeatably. This role bridges product intent, data science and engineering execution, accelerating time-to-value while ensuring AI solutions are secure, supportable, efficient and aligned to enterprise standards and platforms.

Reporting & Stakeholders

Solid line into FDT Tech Lead. Strong day-to-day partnership with Product Owner(s), business SMEs, data/ML colleagues and delivery pod leads on scope, prioritisation and trade-offs. Close collaboration with platform teams (e.g., AI platform / Data platform), architecture, and operations/support teams to ensure production readiness. Engage with tooling/platform owners to provide structured feedback and reusable patterns from field delivery.

Scope & Authority

Accountable for shaping, building and delivering advanced AI prototypes and end-to-end thin-slice AI solutions close to users, including model selection, GenAI solution design, evaluation, safety controls, and efficiency optimisation. Owns local technical decisions required to deliver outcomes at pace, within approved architecture patterns, engineering standards, controls and governance frameworks. Authority includes recommending AI approaches and evaluation methods, implementing guardrails and policies-as-code patterns, and escalating material risks, dependencies or control gaps.

Role Description & Core Accountabilities

This role exists because high-impact AI delivery often requires engineers embedded with users to rapidly discover the right AI approach, prove value beyond simple baselines, and then harden and deliver AI capabilities into production pathways. The AI Engineer accelerates learning loops while maintaining engineering discipline, ensuring what’s built can scale, be supported and be reused.

• Advanced modelling and algorithm selection - Choose and implement more complex approaches when needed (deep learning, graph ML, NLP, GenAI, multimodal, optimisation, RL where appropriate). Build prototypes that demonstrate lift over simpler baselines and justify added complexity.
• LLM/GenAI solution design (if in scope) - Define prompting strategy, tool/function calling, RAG design (chunking, embeddings, retrieval evaluation), context window management and guardrails. Manage hallucination risk through grounding, citations, fallback behaviours and robust evaluation harnesses.
• Evaluation frameworks & AI quality - Create robust offline and online evaluation (golden datasets, human-in-the-loop review, red teaming, safety testing). Define model confidence, uncertainty handling and error taxonomies to drive measurable quality improvements.
• Model efficiency & production readiness - Optimise for latency and cost (distillation, quantisation, caching, batching, model selection). Ensure reproducibility and handover-ready artefacts (model cards, evaluation reports, reproducible pipelines).
• AI safety, security, and controls (technical depth) - Address prompt injection and data exfiltration risks, privacy constraints, and secure use of embeddings/vector stores. Help define guardrails and policies-as-code patterns with engineering and risk partners.
• Deployment partnership - Package models, prompts and retrieval components with MLOps/engineering; define batch vs real-time serving patterns where relevant. Specify monitoring (quality, drift, safety signals, cost/latency) and operational thresholds; support production handover.
• Reusable components - Build shared libraries, templates, evaluation harnesses and patterns that multiple squads can adopt. Prefer reuse over bespoke and contribute reference implementations back to communities of practice.
• Technical leadership - Coach the squad on best practices; align with central AI standards; review architecture choices and implementation quality. Keep delivery moving with crisp weekly milestones and transparent trade-offs.

Calibration Rationale

This is a hands-on, user-adjacent AI engineering role optimised for speed-to-value and end-to-end delivery of advanced AI solutions. Core capabilities around engineering discipline, customer centricity, evaluation rigour, integration delivery and pragmatic decision-making are Essential. Broader strategic leadership and talent stewardship are Desirable, as the role contributes to scaling patterns and mentoring but does not typically own enterprise-wide strategy.

Experience — Required

• Proven experience delivering AI/ML solutions into production pathways, including reproducible training/inference and clear handover artefacts.
• Strong ability to rapidly prototype, iterate with users, and then harden AI solutions towards production readiness.
• Experience selecting and implementing advanced modelling approaches (e.g., deep learning/NLP/graph/GenAI) and demonstrating lift over baselines.
• Experience designing robust evaluation frameworks (offline/online), including golden datasets, human-in-the-loop review and clear quality metrics.
• Working knowledge of AI safety and security risks (e.g., prompt injection, data leakage, privacy constraints) and how to implement practical technical mitigations.
• Experience collaborating with product and business stakeholders to shape MVP scope, define acceptance criteria, and manage trade-offs.
• Comfortable operating in ambiguous environments with shifting requirements, while maintaining delivery discipline and transparency.

Experience — Preferred

• Experience delivering within regulated/controlled environments (risk, audit, security, data handling) and navigating governance pragmatically.
• Experience with GenAI architectures including retrieval evaluation, grounding/citation patterns, and guardrails/fallback design.
• Experience optimising models for latency/cost (quantisation, distillation, caching, batching) and setting cost/performance budgets.
• Experience contributing reusable accelerators (shared libraries, templates, evaluation harnesses) and coaching other engineers.
• Experience working in Value Streams / product-led delivery, transformation pods, or embedded engineering models.

Behavioural Skills

• User-obsessed builder; stays close to users, validates fast, and focuses on measurable outcomes over perfect paperwork.
• Pragmatic engineering discipline; moves quickly without skipping fundamentals (security, testing, operability, controls).
• End-to-end ownership mindset; takes accountability from discovery through build, release and handover; doesn’t throw work “over the wall”.
• Evidence-led decision-making; insists on baselines, measurable lift, and fit-for-purpose evaluation before scaling complexity.
• Clear communicator under pressure; explains options, trade-offs and risks in plain language; keeps stakeholders aligned.
• Collaborative and inclusive; works well across roles and backgrounds; welcomes different perspectives to get to better solutions faster.
• Proactive risk management; identifies safety, security, dependency and control risks early; escalates with options, not surprises.

Derailers / Red Flags

• Adds model complexity without proving measurable lift over simpler baselines.
• Builds AI prototypes that cannot be industrialised, creating rework or hidden technical debt.
• Ignores safety, security, controls or operational needs in the name of speed; pushes fragile AI into production pathways.
• Treats evaluation as an afterthought; lacks clear quality metrics, error taxonomy, or reproducible evidence.
• Over-builds bespoke components where reusable platforms/patterns exist.
• Fails to manage dependencies or communicate risks early, causing delivery churn and stakeholder surprises.
• Focuses on delivery completion without ensuring supportability, monitoring and clear ownership.

Year 1 Outcomes (Scale & Embed)

• Consistent delivery of user-validated advanced AI prototypes that convert into production-viable thin-slice solutions with clear value realised.
• Advanced AI architecture delivered (e.g., RAG/agentic workflow) with measurable lift vs baseline and clear acceptance criteria met.
• Robust evaluation harness and safety tests implemented (golden datasets, red teaming, human review loops), with clear quality and risk controls.
• Performance/cost optimisation plan executed, demonstrating improved latency and/or reduced run cost while maintaining quality.
• Reusable patterns/components contributed back to the Value Stream/platform communities, reducing duplication across teams.
• Strong production readiness discipline established (reproducibility, monitoring, runbooks, handover), with reduced post-release incidents and rework.
• Clear, structured feedback provided to platform/tooling teams, influencing roadmap improvements based on field delivery realities.
• Recognised as a dependable embedded AI engineering partner who unblocks delivery and raises overall AI engineering maturity.