An autonomous brokerage is only as good as the models it trusts. You will close the gap between promising model behavior and production behavior, taking models and improvements out of notebooks and into systems that act correctly without review in a regulated industry. Every point of reliability you add is a category of work that no longer needs a human.
What you'll do
- Improve how our systems perform in production through better prompts, model selection, retrieval, fine-tuning, and system design.
- Own the workflows that let us test, measure, and ship model changes safely.
- Build evaluation systems tied to what matters for the business: bound policies, compliance, and customer experience.
- Turn ambiguous failures in production into concrete model and system improvements.
- Expand what the system handles on its own by proving, with measurement, where it can be trusted.
What we're looking for
- You have shipped ML or LLM systems into production and have seen what breaks outside the demo.
- You are strong in Python and comfortable owning the surrounding engineering, not just the modeling layer.
- You have good judgment about tradeoffs across quality, latency, cost, and operational complexity.
- You thrive in an iterative loop of ship, measure, debug, and improve.
- You care about outcomes in the real system, not benchmark numbers in isolation.
Nice to have
- Experience with model evaluation, routing, fine-tuning, or retrieval in production.
- Experience building internal ML platforms or experimentation infrastructure.
- Experience in insurance, fintech, or other domains where correctness and auditability matter.