Conversation design at Flo Health

Designing AI-powered health conversations for 75 million users

The challenge

Flo is a women's health app used by 75 million people. Building conversational AI at that scale means constantly navigating one tension: how do you make a chatbot feel useful without crossing into medical diagnostic territory?

I worked on two projects that tackled this from different angles.

Symptom Checker — architecting conversations from the ground up

When a user logged a symptom, they could enter a chatbot conversation designed to help them understand what might be going on, without ever diagnosing them.

Working closely with Flo's senior medical advisor, I mapped what could be said, what couldn't, and where the conversation needed to redirect to a healthcare professional. Inside those guardrails, I designed the full branching conversation architecture across three symptom categories — cramps, discharge, and cycles — including logic for how the experience changed for returning users.

AskFlo chip logic — teaching an LLM editorial judgment

AskFlo, Flo's AI health assistant, uses "chips", short suggested follow-up questions — to guide users deeper into a conversation. We couldn't let the LLM generate those freely; in a health context, the wrong suggested question could alarm a user or undermine trust.

I built the rule system governing how the LLM generates chips — covering sensitive topics, symptom-focused queries, tone, and repetition — written in the few-shot format used to train and evaluate LLM behavior. I also optimized the existing chip library using performance data, identifying what users actually clicked and why.

Role Senior Content Designer, Flo Health Collaborators: ML Engineers, Senior Medical Advisor, Product Design