AI Engineer | Agentic Systems & LLM Platform Central London (In person) | Permanent We're hiring on behalf of an AI-first analytics startup whose product runs on a production multi-agent system: Anthropic, OpenAI and Perplexity models orchestrated on AWS, and exposed to customers through the company's own MCP server. This is a real AI engineering role, not calling an API from a web app. The platform already has real users. The job is deciding what good looks like and building the agents, evals and infrastructure that get it there, while setting the technical bar for the team. What you'll do - Design and evolve multi-agent systems: agent topologies, tool calling, memory and context management, and the agent harness itself - Extend the company's MCP server with new tools and capabilities - Build the eval and observability layer: golden datasets, regression suites, LLM-as-judge, and cost and latency tracking per model and client - Own the AWS infrastructure behind it (ECS Fargate, Lambda, Step Functions, S3, DynamoDB, Bedrock), plus CI/CD and infrastructure as code - Guard against prompt injection, PII leakage and cross-tenant data access - Set technical direction, make build-vs-buy calls and mentor other engineers What we're looking for - 3–5 years of software engineering, including 2+ years shipping LLM or GenAI systems in production - Strong Python, and hands-on experience with an agent framework (LangGraph, Strands, CrewAI, OpenAI Agents SDK, Claude Agent SDK or similar) - Production use of LLM APIs, including tool calling and structured outputs, plus RAG (embeddings, vector stores, reranking) - Experience building LLM evals and observability - Solid AWS, and system design for distributed, event-driven systems - A bonus: MCP servers or tools, multi-tenant AI systems, cost optimisation across LLM providers, open-source AI contributions - Right to work in the UK is a must If you use AI agents in your own workflow and have strong opinions on how agentic systems should be built, we'd like to hear from you