What the work involves

This is an engineering seat on a mathematics research initiative, not an annotation queue. You will be designing and deploying agentic workflows that carry out multi-hop reasoning, generate candidate hypotheses, and then validate them through tool calls and structured checks. Day to day, that means sitting with mathematical researchers, taking dense proof sketches or research notes, and turning them into structured specifications an orchestrated agent can actually execute — then building the tooling, memory layers, and state management that keep those runs reproducible.

The second half of the job is evaluation. You are expected to instrument your own pipelines: trace logic across execution steps, find where a chain silently drifted from the mathematics, and iterate on prompts, tool boundaries, or graph structure rather than patching symptoms. Reusable architectural frameworks are an explicit deliverable, so work that only functions for one prompt family will not land.

What the screen looks for

  • Production evidence of multi-step agentic systems — LangGraph, AutoGen, CrewAI, or an orchestration layer you wrote yourself. Expect follow-ups on state, retries, and failure handling, not framework name-dropping.
  • Real STEM grounding: mathematics, physics, CS theory, or quantitative finance, at a level where you can read a proof and identify which steps are mechanizable.
  • Advanced Python, structured output generation, tool integration, and a habit of measuring agent reliability rather than eyeballing it.
  • Comfort as the translation layer between a subject-matter expert and a modular software system.

Logistics

Fully remote contractor assignment through Turing, stated as four weeks in duration and 40 hrs/week, with a required overlap of at least four hours with PST. No medical or paid leave. Pay is undisclosed on this listing; Turing typically negotiates a rate per engagement, and nothing here should be read as a guaranteed band. Be ready to state your PST overlap window and your availability start date in concrete terms during the screen.