Staff AI Engineer - AI Transformation

The Role

Multiverse is the UK's largest apprenticeship provider and its first EdTech unicorn. The current state of AI presents a huge opportunity to reshape the future of education and workforce development. Multiverse is in a uniquely strong position to do that, and getting it right has implications beyond the company: for the UK tech sector and the broader economy.

The AI Transformation team exists to make that real, starting with Multiverse itself. This is not a team that bolts AI onto the edges of the business or ships a handful of internal productivity tools. The mandate is bigger: to rebuild how the company actually works, function by function, and to establish the engineering practices that make Multiverse an AI-first company from the core out.

That work matters twice over. Get it right inside Multiverse and we move faster, serve learners better, and operate at a level few organisations can match. But Multiverse also exists to build the workforce that every other company is reaching for. The way we transform ourselves becomes the standard we set for everyone else. You are not just changing one company, you are building the blueprint others will follow.

The team is one small, focused squad, accountable for outcomes end to end. You work closely with the wider engineering org building Multiverse's customer-facing product, and alongside the teams whose work you are helping to reinvent. The structure is flat and fast. No shared queues, no bureaucratic overhead between having an idea and shipping it.

Whilst we are building something entirely new, Multiverse has an established product, existing infrastructure, and engineering teams in London and Berlin. You need to be as comfortable integrating existing systems and working across team boundaries as you are building new ones from scratch.

What You Will Do

Own the architecture of our internal agentic operating system. The team's work spans the full surface of how Multiverse operates. You own the technical architecture of our agentic operating system: the agent orchestration, context strategy, tool integrations, evaluation framework, and production operation. Your design decisions shape what is possible for human and AI teams at Multiverse

Ship production AI agent systems. This is a building role. You write code, review code, and own the quality of what goes to production. You will personally build and deliver significant agent systems. On a squad this size, nobody leads from a whiteboard.

Design multi-agent coordination. Task decomposition across agents, handoff protocols, shared state management, orchestration logic. You know the difference between agents that genuinely coordinate and agents that run sequentially and hope for the best. You design the patterns that make multi-agent systems reliable.

Build the evaluation and quality infrastructure. Automated eval pipelines, human-in-the-loop review systems, regression testing for prompt changes, domain-specific quality metrics. You treat evaluation as a first-class engineering concern and build the systems that make it possible at scale.

Drive cost engineering. Token economics, caching strategies, model routing, prompt optimisation. The cost profile of production AI systems requires active engineering attention, and you build the cost awareness and tooling into the architecture rather than bolting it on later.

Build the integration layer that makes existing Multiverse systems agent-accessible. APIs, MCPs, shared data contracts, and the tooling that connects agents to the platform, content systems, and the tools the company runs on. This means building real working relationships with engineering teams across London and designing interfaces that serve both sides well.

Set the standard. You define patterns for prompt management, retrieval, guardrails, and testing that the wider team and eventually the whole organisation adopts — and that, in time, shape how the companies who learn from Multiverse do this too. You do this through code, documentation, and architectural decisions, not through mandates.

Mentor the team. Code review, architectural guidance, pairing on the hardest problems. You are not a line manager, but your technical leadership directly shapes the growth of the engineers around you.

What We Are Looking For

Production AI Agent Engineering

You have shipped multi-agent systems or complex AI products to real users. You understand the engineering challenges that make agent systems a distinct discipline:

• Context management. Designing what enters the context window and what stays out. Retrieval strategies, chunking, conversation memory, summarisation, and the cost/quality trade-offs of each. You have made these decisions in production and seen the consequences.
• Model selection and routing. Choosing the right model for each task based on capability, latency, cost, and reliability. Building routing logic that matches work to the appropriate model rather than defaulting to one.
• Cost engineering. Token economics, caching, prompt optimisation, batching. You know the difference between a prototype that works and a production system that works at sustainable cost. You have built systems where cost was an engineering constraint, not someone else's problem.
• Tool use and agent augmentation. Designing what capabilities agents can reach: tool descriptions that models use reliably, failure handling, MCPs or equivalent interfaces. You understand that the quality of the tool layer determines whether agents are useful or fragile.
• Multi-agent coordination. Task decomposition across agents, handoff protocols, shared state, orchestration logic. You have built systems where multiple agents work together within a product domain and understand the architectural patterns that make coordination reliable.
• Evaluation and quality. Building eval frameworks for AI output: accuracy, helpfulness, safety, domain-specific criteria. Automated pipelines and human-in-the-loop review. You would not ship an agent system without a quality baseline.

Product Thinking and Entrepreneurial Instinct

On a small squad there is no gap between product thinking and engineering. You own the problem from user need to production system. You can sit with the people whose work you are transforming, understand their workflow, identify the highest-value intervention, and build it without waiting for a product manager to write a spec.

You have either built something yourself (a product, a startup, a project with real users) or operated with that founder mindset inside a larger organisation. You understand that speed matters and that shipping something useful beats polishing something theoretical.

AI-Native Engineering

You build with Claude Code daily. You set context and constraints before generating code. You review AI output critically. You augment the tool with skills, system prompts, and domain context to make it effective. This is how the team works, and you help define what good looks like.

Full-Stack Delivery

You work across the stack: LLM integration, backend services, data pipelines, and enough frontend to ship end to end. The boundaries between these layers dissolve in agent systems, and so should your willingness to work across them.

Communication

You can explain technical strategy to a CPO, walk a product manager through a cost trade-off, and give direct feedback in code review. You represent the team's technical approach in cross-functional forums with product, design, learning design, compliance, and other engineering teams. You document decisions, not just code.

What Would Set You Apart

• Experience in EdTech, regulated content, or domains where AI output quality has compliance or accreditation implications
• Background as a founding engineer or technical co-founder
• Published thinking or external contributions in AI engineering (talks, writing, open source)
• Experience designing platform layers that other teams build on
• Practical experience with MCP (Model Context Protocol) or equivalent agent integration standards

What We Are Not Looking For

• Pure ML research without production engineering experience. We need builders
• Narrow specialism. This team works across the full stack of an AI product. If you only do infrastructure, or only do model training, or only do frontend, this is the wrong fit
• People who need a detailed spec, a sprint plan, and a standup before they can write a line of code. We ship fast and iterate
• Candidates whose experience is limited to wrapping LLM APIs in thin application layers. We need depth in agent architecture, context strategy, tool design, and multi-agent coordination
• Engineers who optimise for technical elegance over user outcomes. The architecture serves the product