The problem: During cardiac arrests, someone (often a nurse) has to be the "recorder" - basically frantically scribbling notes while everyone else tries to save a life. It's chaotic, error-prone, and honestly, an unproductive use of human brainpower during critical moments.

My role: Lead Product Designer & Clinical Translator

How we tackled it:

• Translated complex clinical workflows for non-clinical engineers and product managers, explaining ACLS protocols, room dynamics, stakeholder impact, and evidence-based guidelines

• Designed voice-to-text system achieving <2s latency and >95% accuracy that could handle medical chaos (turns out "epi" sounds different when someone's stressed)

• Engineered clinical prompts and training data for AI coaching module, ensuring alignment with AHA ACLS guidelines and real-world emergency workflow requirements

• Led UI/UX design and facilitated technical constraint discussions to balance clinical accuracy with technical feasibility

• Built in AHA ACLS protocol reminders because guidelines exist for a reason and can coach clinicians on next steps

• Created post-event reports that actually help teams improve instead of just checking compliance boxes

The cross-functional collaboration: Worked cross-functionally with developers on keyword spotting and fuzzy term mapping. Basically taught the AI that when someone yells "push another epi!" it means epinephrine, not a casual suggestion. Managed multiple demo iterations and testing cycles.

The win: 3rd place + People's Choice Award at Out Of Pocket Health's Hackathon. But more importantly, the demo worked flawlessly in front of 300+ healthcare leaders, including VCs from a16z and Rock Health, plus startup executives from Ambience and Suki.

Why it matters: We built the exact differentiator that top AI scribe companies desperately need to compete with Epic. While they focus on routine documentation and RCMs, we spotted the high-stakes opportunity they missed - emergency scenarios where documentation errors cost hospitals millions.

Hospitals spend $12.8 million annually just on nursing labor for code documentation, plus malpractice exposure of $250,000 to $5 million per settlement often stemming from documentation failures. We built scalable AI that addresses hundreds of millions in addressable costs and created defensible IP around medical crisis protocols.

What I learned: The biggest challenge wasn't the technical AI piece - it was getting non-clinical stakeholders to understand the workflow complexity. Next time, I'd create visual workflow maps earlier in the process to accelerate buy-in and technical requirements gathering.