Applied AI Engineer - Commodities trading - Up to £140k base plus a high bonus

Every trading floor is full of workflows that quietly eat hours: a trader re-reading the same contract clauses, an analyst manually stitching together data from five systems, a knowledge base nobody trusts enough to rely on. AI can fix a lot of that - but only if someone actually builds the bridge between "AI is powerful" and "AI works reliably for this specific desk, this specific document, this specific decision." That bridge is this role. This is a global commodities trading firm looking for someone to sit inside their Front Office tech team and turn approved AI tooling into something traders, analysts, and operators actually trust and use. You won't be researching AI in the abstract - you'll be in the weeds with the people who need it, figuring out where a workflow is slow or error-prone, configuring the agents, knowledge bases, and retrieval systems to fix it, and then making sure the output is accurate enough to stand behind. That means real context engineering, writing the instructions, examples, and guardrails that make an AI system dependable — and real evaluation, testing outputs against what a subject-matter expert would actually accept. Then deploying these solutions into production. It's a role that lives at the intersection of technical depth and plain communication. One day you're debugging chunking strategy on a document-ingestion pipeline or wiring up an MCP tool; the next you're standing in front of a commercial team explaining, in language that has nothing to do with embeddings, what the tool can and can't be trusted to do. You'll build the reusable playbooks and demos that let good AI habits spread across the firm rather than living in one person's head. You’ll need to be someone who’s worked within the commodities space as this is their first hire in London. Also, deployed LLM agentic solutions into production. At least a couple of years of hands-on experience actually building with generative AI: prompting, retrieval, structured outputs, agentic workflows, evaluation. You should be comfortable in Python, familiar with the modern AI toolkit (LangChain, LangGraph, Pydantic, vector search, MCP), and just as comfortable explaining risk and hallucination trade-offs to a non-technical stakeholder as you are debugging them yourself. Above all, you should care about the difference between AI that looks impressive in a demo and AI that a trader will actually rely on when real money is on the line. This position requires you to be in the office 4 days per week. No up-to-date CV required at this stage.