Software Engineer

### Who you are
- 2 - 5 years of experience building software professionally, with strong programming fundamentals (we work mostly in Python)
- Proficiency with LLM APIs, agent frameworks, and modern retrieval patterns. You stay current as the field moves. We don't need classical ML expertise - we need someone on the frontier of applied AI
- Strong track record of shipping high-impact work in ambiguity. You take a vague problem, define what "good" looks like, ship a v0, and iterate
- Experience shipping user-facing products, either backend or full-stack
- Drawn to the question "how do we know this is actually better?" - excited to build the evaluation infrastructure that answers it rigorously
- Curiosity over domain experience. No insurance background needed - you learn fast and care about understanding the problem space deeply
- Strong written and spoken English

### What the job involves
- Our AI team is small, talent-dense, and operates with real ownership. We maintain open-source projects with thousands of stars and publish work recognized across the AI ecosystem. The problems we work on are active research areas where no one has great answers yet
- You'll own the full feedback loop: diagnose where the product falls short, rigorously analyze root causes, build evaluations, and ship improvements that customers feel
- The work spans agent systems, evaluation and data quality, and production engineering in roughly equal measure
- You'll work directly with customers when it helps you understand the problem deeply - we believe engineers should have a clear line of sight to the people using what they build
- Design and refine tool-calling patterns and agent orchestration to handle multi-step research tasks that requires reasoning across thousands of related documents
- Build ingestion and normalization systems for complex structured documents at scale - millions of filings where formats vary and each version builds on the last
- Build and iterate on reinforcement fine-tuning harnesses, including training graders for research tasks with non-verifiable reward
- Develop tooling to measure, profile, and optimize agent pipeline performance - latency, cost, and quality across millions of calls
- Run experiments end-to-end: design, analyze, and translate into clear product recommendations
- You'll work on genuinely hard, unsolved problems at the intersection of AI systems and a massive real-world data domain
- You'll ship to production regularly, talk to customers, and see your work change how people make consequential decisions
- The team is small enough that your judgment directly shapes the product - and we're building it to be best in class