AI engineer for agent platforms

wingmaite captures the knowledge companies never write down: who decides what, how work actually gets done, and why decisions were made. Our AI agents use that knowledge to answer questions and get work done. We're looking for a Staff engineer to own the technical quality and direction of our agent platform. You'll build features end to end, from the agent into the backend. You won't manage people. You'll work closely with our CTO as a technical sparring partner.

• Decide how our agents plan, use tools and handle multi-step work, and where people need to step in.
• Build reusable components so each new use case doesn't need its own system.
• Build how that knowledge reaches the agents, together with the people who decide what it should contain.
• Agents should be correct, cite their sources, show people only what they're allowed to see, and know when not to act.
• Build evals, monitor quality and cost, and keep our model choices up to date.
• Build features from the agent through to the backend yourself.
• Set the patterns others build on, document decisions, keep specs current, and review designs with engineering, product and design.

• You've shipped multi-step, tool-using LLM agents to production (LangGraph or similar)
• You know when sub-agents help, how to build good evals, and why retrieval fails
• Fluent in Python and productive in TypeScript/Node
• PostgreSQL and data modeling
• You use AI coding tools such as Claude Code every day
• You write clear specs and can explain trade-offs to people who aren't engineers
• Fluent English (C1–C2)
• You've set technical direction beyond your own work, and others build on your patterns
• You start from what users need and what could go wrong, then choose the tools
• You care whether the system still makes sense in a year
• You're happy to do backend work when a feature needs it
• You want influence through your work, not through a title
• You like working with designers and product people
• You enjoy technical debate and change your mind when the other argument is better
• Будет плюсом: NestJS is a plus; You've connected agents to tools and data through MCP; You've worked with knowledge graphs or other ways of modeling how people, work and information relate; You've run models in production on AWS Bedrock or Vertex AI; You've traced and debugged agents with LLM observability tools such as Langfuse; You speak German

• Flexible remote setup
• Learning & development opportunities
• Generous vacation & workation budget
• Technical ownership of the platform at the core of wingmaite's product