FloodReady Houston

Your address, your route, your language: know the real flood risk before the water rises.

Not official guidance. In an emergency, call 911.

Inspiration

Houston floods, and our tools are built around gauges and 2007 maps, not people.

• Harvey flooded 154,170 Harris County homes; only 36% of the flooded homes were insured, and nearly half were outside the 100- and 500-year floodplains (Community Impact).
• 77% of Texas flood deaths (1959-2008) came from driving into water (UTHealth).
• Updated rainfall data show rates up more than 30% in Harris County, but the flood maps in force predate them; the new FEMA maps are still drafts (ReduceFlooding).

We catalogued 52 concrete Houston flood problems with sources and built one app around them.

What it does

• Know your risk: an address shows the FEMA zone beside "flooded in Harvey" (357,241 citywide polygons), elevation, gauges, reservoir area, evacuation zone and storm-surge depth, flood history, and ZIP flood-insurance claims.
• Drive Safe: 128 flood-prone crossings, a caution / avoid route check (never a plain "safe"), community reports with "Still there / Cleared" votes, NWS warning polygons, and a suggested detour with fewer flood-prone spots.
• Ask FloodReady (agentic AI): a cited, multilingual assistant (LangGraph + RAG) with 16 live data tools. A "How I answered" trace shows the sources and tools it used.
• Voice mode: full-duplex Gemini Live that already knows your address, with barge-in and captions.
• After a flood: FEMA declaration banner, NFIP proof-of-loss calculator, SBA loans, appeal steps, scam warnings, floodwater and electrical safety, mold, and "waiting for rescue" and "no home" cards.
• Vulnerable households: dialysis centers show their flood zone and Harvey flooding; households without a car or with medical needs get "get help to evacuate now" (2-1-1, STEAR, 911).
• Works offline: an offline safety card (911/311/211, your zone and elevation, household needs, kit reminders).
• Reach everyone: 5 languages (English, Spanish, Vietnamese, Chinese, Arabic), SMS, Web Push alerts, installable PWA, easy-read mode.
• One labelled demo storm so judges can see storm behavior, always tagged "SIMULATED - demo only."
• Everything works without API keys, using fallbacks.

How we built it

React + Vite + TypeScript + Tailwind + Leaflet PWA with a service worker and Web Push. FastAPI backend (Python 3.13) over USGS, NOAA, NWS, FEMA, OpenFEMA, Census, OSRM and Open-Meteo, plus bundled datasets. AI: Gemini for plans, answers and embeddings, Gemini Live for voice, ElevenLabs for speech; BM25 search and templates are the no-key fallback. Backboard household memory and Tiger Data (TimescaleDB + pgvector) are integrated and turned on by configuration. Deploy-ready with Docker Compose + Caddy for Vultr. Built with parallel Claude Code agents.

Challenges we ran into

• The live roadway flood-sensor feed is restricted, so we built our own crossing dataset from public sources.
• Slow government APIs: aggressive caching and pre-extracted Harv
• Honest geography: Harvey data covers only the City of Houston, so elsewhere we say "no Harvey data" instead of a false "not flooded." - Gemini free-tier quota ran out overnight, so we built key and mks.
• Our own audit caught real safety bugs (one report flipping a route to "avoid", an SMS distress text without 911). All were fixed and re-tested.
  ## Accomplishments that we're proud of - A strict problem-by-problem audit: the app reduces harm for **3s on 13, and we say plainly that 6 are beyond any app.
• 243 backend tests passing, AI eval 38/40 (95%), citation eval 15/15, voice response median 1.4 s. - Accessibility: 0 axe-core violations across 48 view, width and ## What we learned Flood risk is hyperlocal: a gauge or a map zone doesn't tell you whether your street or commute is safe. Plain language, the right language, and a clear "avoid this route" matter more than more data. We dropped claims than ship them. ## What's next Native app for reliable push, route warnings while driving, live SMS, native-speaker review of translations, and Harvey-style coverage beyond the City of Houston.

Built With

• backboard
• caddy
• docker
• elevenlabs
• fastapi
• gemini
• langgraph
• leaflet.js
• noaa
• openstreetmap
• pwa
• python
• react
• tailwindcss
• timescaledb
• typescript
• usgs
• vite
• vultr