Back to jobsDesign Agentic Workflows & Problem Decomposition: Architect end-to-end workflows using LangGraph (or similar), breaking complex problems into composable steps, states, and policies. Translate high-level business problems into structured agentic solutions, defining workflows, components, and trade-offs across multiple design options.
Build Skills, Tools, and Agents: Implement reusable skills, tools, and agents, and compose them into workflows that are maintainable and testable. Demonstrate strong judgement on when to build a skill vs. a tool vs. an agent vs. a workflow, based on problem decomposition, system design trade-offs, and long-term maintainability—balancing accuracy, complexity, observability, and long-term reuse.
Model Strategy, Integration & Performance Optimization: Integrate and operate multiple LLM providers, optimising for latency, cost, and output quality. Optimize prompt engineering, token usage, chunking strategies, and inference efficiency across the platform. Design typed schemas (e.g., Pydantic models) for reliable data flows, tool contracts, and structured outputs.
Evaluations, Quality & Observability: Define and implement evaluation strategies (offline and online), datasets, metrics, and guardrails to validate agent behaviour and measure regression over time. Implement logging, tracing, and monitoring for agentic systems using tools like Langfuse or LangSmith.
Cross-functional Collaboration: Partner with data scientists, engineers, and product teams to translate requirements into robust agentic capabilities and deliver measurable value. Support colleagues and junior engineers through mentorship, code review, and technical guidance. Guide the team in building robust evaluation practices and establishing strong quality standards. Use AI assistants (including Claude Code) in day-to-day work for prototyping, refactoring, testing, and documentation—while applying strong engineering judgement.
