Guarani: AI-Powered Rainforest Defense System

Real-time acoustic surveillance powered by Gemini native audio analysis and ChromaDB RAG to detect illegal logging and poaching in the Amazon rainforest.

Comment

Inspiration The Amazon Rainforest covers millions of hectares, making satellite monitoring insufficient to stop illegal deforestation in real time. We wanted to listen to the forest directly—detecting chainsaws, gunshots, and boat engines instantly before irreversible destruction occurs.

What it does Guarani transforms raw acoustic telemetry from rainforest IoT sensors into actionable tactical intelligence. Key features include:

  • Real-Time Audio Analysis: Direct processing of raw bioacoustic audio files via Gemini API.
  • Interactive Tactical HUD: Leaflet GIS mapping with live sensor locations and threat severity indicators.
  • Precision Waveform Synchronization: Wavesurfer.js highlights the exact timestamps where illegal activity occurred based on Gemini's structured JSON output.
  • Semantic RAG Search: ChromaDB vector memory allowing operators to query past incidents in natural language (e.g., "Find chainsaw incidents near Mamirauá").

How we built it

  • Backend: Built with Python 3, FastAPI, Uvicorn, SQLAlchemy, and SQLite.
  • AI & Audio Engine: Google Gemini 3.5 Flash (via Google GenAI SDK) directly ingests raw audio without intermediate transcription pipelines, utilizing forced response_schema for deterministic JSON timestamp extraction.
  • Vector Engine & RAG: ChromaDB stores alert embeddings with real vector similarity score filtering (>40% threshold) to prevent false positives.
  • Frontend HUD: Built with Tailwind CSS, Vanilla JS, Leaflet.js (CartoDB Dark Matter), Wavesurfer.js, and Chart.js.

Challenges we ran into Extracting precise timestamps from variable-length forest audio feeds without dropping acoustic context, and configuring ChromaDB vector distance thresholding to eliminate false matches from the intelligence feed.

Accomplishments that we're proud of Building a military-grade command center interface that turns raw ambient noise into actionable, geo-located defense intelligence using native Gemini audio capabilities.

What we learned Gemini’s native multimodal capabilities eliminate the need for complex audio preprocessing or transcription pipelines, drastically simplifying real-time bioacoustic classification.

What's next for Guarani

  • Direct integration with LoRaWAN IoT sensors for field deployment in remote Amazon sectors.
  • Automated alert forwarding to environmental defense agencies (IBAMA / ICMBio).
  • Multi-sensor acoustic triangulation for precise sound source positioning.

Built With

Share this project:

Updates