Build automations with judgment in them, and the guardrails that make them safe.
About the Role
Build the automation layer that replaces repetitive judgment work. You will design workflows, wire the tools, and build the evaluation and guardrails that let them run unattended.
What You'll Do
Design and build automated workflows with AI decision steps
Build tool integrations that are idempotent and safe to retry
Implement evaluation harnesses so quality is measured, not assumed
Design approval gates for anything consequential
Monitor cost per task and optimise model routing
Debug automations that fail in ways nobody predicted
What You'll Bring
2+ years building backend services or integrations
Strong Python or JavaScript and API fluency
Practical LLM experience: prompting, structured outputs, function calling
Understanding of idempotency, retries and failure modes
Instinct for what should never be automated
Nice to Have
LangGraph, Model Context Protocol or similar orchestration
Experience with vector databases and retrieval
n8n, Temporal or workflow-engine experience
What You'd Build
These aren't hypothetical projects, they're live products you can try before your first interview.
