01 — Floodline: Building AI flood guidance with New Yorkers

The opportunity

Flood preparedness and recovery information in New York City is spread across agency websites, technical maps, PDFs, community organizations, and emergency resources. The correct guidance can depend on a resident’s address, housing situation, flood type, and whether the event is approaching, currently happening, or already over.

My role

I led product development for Floodline’s AI guidance experience, defining its core user journeys, answer flows, product requirements, content structure, escalation logic, and testing process. The service is designed for renters, homeowners, and small businesses that needed practical, locally relevant actions rather than another collection of links.

What I worked on

I helped move the product from an experimental chatbot toward a structured resident-guidance service that could be tested, evaluated, and continuously improved. I organized more than 100 local sources, maps, resident stories, images, and community resources into reusable knowledge workflows.

I also designed a taxonomy, scoring rubric, and pass/fail testing process covering source grounding, hallucination risk, local relevance, coverage, and actionability across hundreds of flood scenarios. Research with more than 500 residents at 15+ community events informed product, content, and UX decisions and supported pilot work with five neighborhood partners.

This project is still currently ongoing and expanding across NYC.

What I learned

Public-interest AI should focus on making disparate information incredibly easy for anyone to answer. Success for AI in this space should be measured by whether it helps someone take the right next step.