AI · 1 opening · Full-time
Senior AI Engineer
Own the model layer: the agent loop, the ICP gate, tool-calling belts, and the evals that keep a model between a prompt and a person honest.
About the role
Five specialists and a coordinator run every workspace's outbound motion. One of them is the only thing in the product allowed to send, and every model output that could reach a person passes a compliance gate first. You will own how those models are prompted, tooled, measured, and constrained.
The stack is model-agnostic by design, routing across frontier models, with tool allowlists enforced in code rather than asked for in a prompt.
You will
- Design and instrument the agent loop: planning, delegation, tool belts, confirmation gates.
- Own precision-critical judgement passes. The ICP gate that decides who becomes a contact.
- Build evals against real transcripts so a prompt change is a measured change.
- Drive model cost down without letting quality follow it.
You have
- Shipped LLM-backed features to production users, not demos.
- Strong opinions about tool-calling, structured output, and where prompts fail.
- The engineering fundamentals to own the code around the model, in TypeScript.
- A habit of measuring before and after.
Not sure this is the one? Read what we shipped.
The changelog carries the arguments as well as the features, which tells you more about the work than a job description can.
Start with sales. $2,000 or $5,000 onboarding, then at least $250 monthly credit funding.