Job Description

Job Summary:

You will build the AI runtime, retrieval, guardrails, and evaluation that make the platform's AI features work in production - with cost, latency, and safety treated as first-class engineering constraints. We build on proven commercial models and win on integration, domain knowledge, and evaluation discipline.

You'll report to the Head of AI Platform Engineering and join a small, senior founding team building the platform from the ground up.

The opportunity: ship AI that customers depend on daily in mission-critical workflows - not demos.

Job Description:

Why This Role Is Different

• Applied, not research. We use proven commercial models and win on integration, retrieval, orchestration and safety validation. Fine-tuning is an optional track, not the goal.
• Domain-first, not tech-first. Domain knowledge, delivered through retrieval and evaluation, is the moat, not model weights. You will invest early in your customers' domain.
• Production-first. We ship AI that customers depend on daily. Demos don't count; reliability, cost and safety do.
• Platform and product. You will build reusable AI capability while enabling products to ship features fast.
• A founding role. As a found
• ing AI hire, you will set the patterns and evaluation discipline the team adopts, and help hire the team.
• Value-based. You will validate with product teams and customers and prioritise by value, building what moves outcomes rather than what is novel.

What You'll Do

AI Runtime and Orchestration

• Build and operate the runtime, tool calling and orchestration for production agent workflows.
• Design model selection and tiering, with cost and latency as first-class constraints.
• Implement and tune retrieval (RAG) and grounding where it measurably improves quality.

Evaluation, Safety and Quality

• Stand up evaluation harnesses and golden sets, and use them to make ship or no-ship calls and to catch regressions and drift.
• Design and implement evaluation frameworks for model and agent quality.
• Build guardrails and safety controls, treating retrieved content and tool outputs as untrusted.

Production Ownership

• Own the production behaviour of what you ship, including latency, cost, reliability and failure modes.
• Partner with product to decide when an agent is, and is not, the right tool.

Delivering Value with Product and Customers

• Work with product teams and customers to validate results and outcomes, prioritising by value.
• Take a value-based development approach: prove value with the smallest useful slice, then scale what works.
• Instrument and measure impact, letting evidence rather than opinion drive what ships.
• Build automated QA, including tests, quality gates and regression checks, into the delivery pipeline.

Building the Team and Ways of Working

• Help build the team by setting evaluation and safety patterns others adopt, raising the bar in interviews and mentoring.
• Foster a culture of technical excellence, customer obsession and pragmatic shipping.
• Partner with product teams across the portfolio who own their own roadmaps.

What You'll Bring

We look at demonstrated experience and scope, not years served, and we don't expect every candidate to tick every box.

• Production AI: You have shipped software backed by LLMs or ML to real production users, beyond demos and notebooks.
• Evaluation over vibes: You reason about evaluation, not just prompting.
• Cost and latency: You treat cost and latency as design constraints from the start.
• Clear communication: You explain your thinking clearly to engineers, product partners and customers, and bring people along.
• Value and pragmatism: You focus on outcomes and value, and make pragmatic calls to ship.

Bonus Points

Any of these strengthen an application; none are required.

• Agents in production: Agent or RAG systems in production, and the failure modes that come with them.
• Routing and cost wins: Model routing or measurable cost reduction.
• Calibrated evaluation: Evaluation harnesses driving decisions; LLM-as-judge calibrated against humans.
• Optimisation track: Fine-tuning or distillation.
• Judgement: Knowing when not to use an agent.
• Languages: Strength in Python or C#/.NET.
• Semantic data: Experience grounding AI in structured, well-modelled business data.
• Architecture: System and platform design, and trade-off judgement such as one-way versus two-way doors.
• Enterprise scale: Experience building or operating enterprise-scale software, and its reliability, security and integration demands.
• AI-assisted engineering: You use AI coding assistants productively and supervise their output with judgement.

New to some of this? If you bring the essentials above and are excited by the problem, we'd still like to hear from you.

#LI-DNI

Worker Type:

Regular

Number of Openings Available:

3