Mid-Level Software Engineer — Refine

Refine is building the referee your research deserves. We've cracked the problem that general-purpose AI can't solve: spotting mathematical errors, empirical specification mistakes, and logical gaps in academic papers at scale. Researchers at 1,000+ institutions—including central banks and financial institutions—trust us to catch what single-pass tools miss.

What You'll Do

• Design and optimize our parallelized multi-agent LLM pipeline, orchestrating domain-tuned workflows that process papers with hours of compute for substance-level review
• Build and improve model routing logic across multiple LLM providers using LiteLLM (we maintain a fork), balancing cost, latency, and accuracy
• Develop backend systems that handle encrypted storage, SOC2 and ISO 27001 compliance, and reliable processing for high-stakes academic feedback
• Collaborate on infrastructure that scales from pilot deployments to institution-wide adoption
• Debug and refine the feedback loop between orchestration logic and LLM outputs to catch real errors academics care about

Who You Are

• 3+ years shipping production software; strong CS fundamentals are the primary filter — we care most about your ability to reason about systems, algorithms, and correctness
• Comfortable with multi-step workflows, async processing, or orchestration (Airflow, Temporal, or similar)
• You care about correctness and can reason about where AI tools fail in practice
• Good-to-have: experience with Python and/or TypeScript (our primary stack), but not a hard requirement — strong fundamentals matter more
• Nice-to-have: experience with LLM inference, cost optimization, or academic/scientific computing
• Nice-to-have: familiarity with compliance-heavy systems or regulated environments