Skip to main content

Whisperways

Noise-aware eVTOL flight corridors for Los Angeles: route by who hears you, not just how far you fly.

Open source (MIT) · TypeScript · Next.js and MapLibre · 28 vitest tests over acoustics, routing, real-city acceptance, and AI-brief parsing · Built for the HTCJ Aviation Futures Innovation Challenge 2026.

The problem

Community acceptance, and noise above all, is the gate on advanced air mobility. The corridor that gets approved is the one a neighborhood can live under, yet planners lack a fast, honest way to compare corridors by human noise cost and to show a community the tradeoff.

Pick two Los Angeles vertiports and Whisperways returns two corridors, the direct route and the quiet route, and shows how many people can hear an eVTOL overhead under each. It runs a population-noise cost raster, real US Census population convolved with a published eVTOL audible-footprint kernel, through Dijkstra to trade flight meters against person-dB. Then Claude drafts a neighborhood-by-neighborhood community impact brief grounded only in the engine numbers, forbidden from inventing anything beyond them. The first-order acoustic model is honestly labeled a planning heuristic, not certification.

Quickstart

Node with pnpm. The vitest suite runs with no API key; the app needs an ANTHROPIC_API_KEY only for the AI community impact brief.

Clone and install:

git clone https://github.com/OrionArchitekton/whisperways
cd whisperways
pnpm install

Command surface

  • pnpm dev: The Next.js app: pick two Los Angeles vertiports on the map, compare the direct and quiet corridors by how many people can hear an eVTOL overhead, and generate the community impact brief.
  • pnpm test: 28 vitest tests across four files: the acoustics model, the population-noise router, real-city acceptance over Los Angeles, and AI-brief parsing.
  • pnpm build: Production build of the Next.js App Router site (React 19, MapLibre GL, deployed on Vercel).
  • /api/brief: The brief route calls Claude with a schema-constrained prompt fed only the engine numbers, so the draft cannot invent any figure the routing and noise engine did not produce.

Why it is different

  • Routes by who hears you, not just distance: Most routers minimize distance or time. Whisperways scores every corridor by human noise cost: a population-noise raster built from real US Census population and a published eVTOL audible-footprint kernel, so the quiet route is the one the fewest people can hear overhead.
  • The model never invents a number: Claude drafts the community impact brief, but only from the engine numbers. The prompt is schema-constrained and fed the routing and noise results, forbidden from inventing anything the engine did not compute, so the brief stays defensible in front of a neighborhood council.
  • Honest about what it is: The first-order acoustic model is labeled a planning heuristic, not certification. Whisperways exists to compare corridors and start a community conversation, not to replace formal acoustic analysis.
  • Grounded in real data, tested against a real city: Population comes from the 2020 US Census, acoustics from a published eVTOL flyover campaign, and the whole pipeline is covered by 28 vitest tests including a real-city acceptance test over Los Angeles.

Links