London - Applied AI Engineer

Role overview

• Build AI-powered features end-to-end: design, prototype, evaluate, ship, and operate. Frontend integration through to production monitoring.
• Design retrieval-augmented generation (RAG) systems over our data: chunking strategies, embedding models, vector store choice, hybrid search, and grounding.
• Build evaluation harnesses that measure what actually matters faithfulness, hallucination rate, latency, cost, instruction-following — and wire them into CI so quality doesn’t regress silently.
• Design agent architectures using tool use / function calling, structured outputs, and multi-step workflows. Plan for failure modes, not just happy paths.
• Own prompt engineering at the system level: versioning, testing, A/B comparison, and the discipline to treat prompts like code.
• Think about safety and reliability: prompt injection, abuse, misuse, and what “behaves predictably under pressure” actually means for our users.
• Manage cost and latency: model selection, caching, batching, and knowing when a smaller model is the right answer.
• Bring the rest of the team along: show colleagues how to think about modern AI, run internal workshops, and help us build a shared understanding of what’s possible and what isn’t.
• Production AI experience. You’ve shipped at least one real feature powered by a large language model or foundation model, and operated it in production. Demos and side projects are great, but production is where the lessons live.
• Strong Python skills and solid software engineering fundamentals: APIs, testing, CI/CD, version control. AI engineering is still engineering.
• Hands-on experience with major LLM provider APIs: including prompting, tool use, function calling, and structured outputs. You understand the trade-offs between providers, models, and open-source alternatives.
• Practical experience with RAG: embeddings, vector stores, retrieval optimisation, and grounding.
• Evaluation discipline. You’ve built or maintained an eval harness and can talk through what you measured and why.
• A pragmatic, product-minded approach. You know when to fine-tune, when to prompt, when to retrieve, and when to use a deterministic rule instead of an LLM.
• Excellent written communication: most of our deep work happens in writing, and explaining AI trade-offs clearly is half the job.