What you'll do

  • Design and build the agent orchestration that drives a Verety investigation — planning what to check, dispatching tool calls to registries, sanctions lists, and media sources, and deciding when enough evidence exists to return a verdict.
  • Own the investigator LLM's prompting, tool-calling, and structured-output contracts — the boundary between "the model thinks this" and "the system can act on this."
  • Build evaluation harnesses that catch regressions before they reach a customer — this is a system whose mistakes are compliance decisions, not just wrong chat replies.
  • Work directly with real coverage gaps: when a jurisdiction's data is thin or a source is unreliable, decide (and build) how the agent should degrade honestly instead of guessing.
  • Pair with the rest of engineering on the graph, scoring, and API layers the agents feed into and read from.

What we're looking for

  • Production experience building systems on top of LLMs — not just prompting a chatbot, but systems with tool use, structured output, and real failure-mode handling.
  • Strong fundamentals in backend engineering: you can design a data model, reason about latency and cost, and write code other engineers can read.
  • Comfortable with ambiguity — a lot of this role is deciding what "good enough evidence" means for a given jurisdiction and encoding that as real logic, not a one-off hack.
  • Able to work independently in a small, remote, early-stage team.

Nice to have

  • Experience with KYC/KYB, AML, fraud, or other compliance/risk domains.
  • Experience with graph data structures or entity-resolution problems.
  • Experience evaluating or red-teaming LLM systems specifically for correctness, not just helpfulness.