Role overview: As an Agent Architect, you'll design and prompt the systems that power our AI assistant Mia's patient interactions. You'll architect multi-agent workflows and translate research into production agents that handle real patient conversations at scale. You'll get a front-row seat to cutting-edge LLM failure modes that few people globally get to observe—understanding why agents break in production and building systems that consistently work for real healthcare interactions. This isn't about writing complex algorithms from scratch or chasing the latest AI trends. Instead, you'll work on problems most AI engineers don't get exposure to: analyzing conversation patterns to understand failure modes, designing evaluation frameworks based on real patient behavior, and methodically debugging production issues that emerge when language models meet messy healthcare workflows. This role is ideal for those who thrive at the intersection of LLM system design, product innovation, and applied AI research—where the challenge isn't building impressive demos, but making cutting-edge AI actually reliable for people who need it. Exercise Overview: Conversational Analysis In this exercise, you’ll analyze 50 anonymized patient-agent conversations. Your goal is to uncover why the agent isn’t achieving desired outcomes, identify key failure patterns using Python/NLP, and recommend specific, testable changes to the conversation flow. You’ll also design metrics to evaluate your proposed solutions in production. We’re looking for clear investigative thinking, pattern recognition, and practical, user-grounded improvements. We hope you enjoy the process and treat it like an opportunity to show us how you actually work and think!