FrontEar

FrontEar was built for one reason: indigenous communities in the Amazon are invisible to the health coordinators who could help them.
Field workers visit these communities regularly, but the data they collect (and the way they do it) — audio notes, clinical observations, supply needs — goes nowhere. There’s no internet, no structure, and no way for coordinators to act on it.
We built two connected apps in 8 hours. The field app works entirely offline: operators record audio anamneses, attach photos, write notes, and tag their GPS location. Google AI Edge ASR transcribes audio on-device. Everything queues locally until the operator reaches connectivity, then syncs automatically to Firebase. From there, Gemini 2.5 Flash structures the raw data into standardized triage records — symptoms, severity, supplies needed, community ID.
The coordinator dashboard receives that data in real time. It shows a live community map with health heatmaps, flags supply shortages before they become emergencies, and includes an AI agent that answers operational questions directly: “Which communities need resupply this week?”, “What did the last Tapauá anamnesis say?”
The stack is Firebase-first with no custom backend, a Turborepo monorepo for speed, and Gemini handling both structured extraction and conversational RAG over field data.
The result: communities that were previously unreachable become operationally visible. Coordinators can prioritize flights, pre-position supplies, and respond to critical cases with actual data behind every decision.

The idea of the project came from an actual experience from our product leader when he was on a mission on local communities at Brazilian Amazon Forest.