Project AngelWare: AI-Enhanced Search and Rescue

Project AngelWare is a hardware-software integration designed to assist firefighters in high-stakes search-and-rescue operations. By combining ESP32-based CSI (Channel State Information) sensing with AI-driven automated dispatch, AngelWare ensures victims can be found and dispatch is notified as quickly as possible.

The Problem

In active fire scenarios, communication is hindered by smoke, noise, and physical exhaustion. When a firefighter discovers a victim, relaying information to medical teams via radio is often slow or impossible, delaying life-saving care.

The Solution

Project AngelWare uses broadband wifi signals to detect the presence of survivors despite low visibility and streamlines the transition from detection to dispatch through a silent, one-touch interface on a wearable display.

Core Features

• CSI Sensing Mesh: Utilizes ESP32-S3 nodes to detect human presence/motion behind walls via signal disruption.
• One-Touch Activation: Firefighters tap the victim's location on a Freenove DSI display powered by a Raspberry Pi.
• Silent AI Dispatch: * Twilio Voice API: Automatically initiates an emergency call to hospitals.
  • ElevenLabs Conversational AI: Converts LLM-generated situational reports (SITREPs) into natural, high-clarity speech for the receiver.
• Edge-to-Cloud Inference: Real-time camera capture processed via OpenCV and streamed over WebSockets to a RunPod GPU instance running DensePose for advanced pose estimation.

---

Tech Stack

Languages

• Python: Backend logic, AI integration, and UI.
• C: ESP32-S3 mesh node firmware.
• PowerShell: Deployment and environment orchestration.

Frameworks & Libraries

• UI/Frontend: Pygame
• Computer Vision: OpenCV (Capture, JPEG encode/decode), Detectron2 + DensePose
• Machine Learning: PyTorch, NumPy
• Communication: WebSockets (Real-time streaming), PySerial (Hardware comms), Requests, python-dotenv

Hardware & Platforms

• Edge Host: Raspberry Pi (Frontend & Display controller)
• Sensing: ESP32-S3 mesh nodes (CSI sensing)
• Display: Freenove 800x480 DSI Touchscreen
• Cloud Inference: RunPod GPU Instances (DensePose server)

Cloud Services & APIs

• Voice & Audio: ElevenLabs (TTS & Conversational AI), Twilio Voice API
• Networking: RunPod Cloud GPU for signal processing. Specs: runpod/pytorch:2.1.0-py3.10-cuda11.8.0-devel-ubuntu22.04 RTX 40901x Memory 62 GB container disk 20 GB
• Mongo DB Atlas Stores Gemma processed ESP32 Mesh signals and presence detection data for future training of the model.
• Google Gemma Formats ESP32 Mesh signals and presence detection data based on an expected training data template for the model.

Storage

• Logging: Local JSONL event logging (tailed by the frontend for real-time motion event updates).

---

Getting Started

Installation

1. Clone the repository:

   git clone [https://github.com/JustinPMoran/lahacks-2026.git](https://github.com/JustinPMoran/lahacks-2026.git)
   cd lahacks-2026

2. Configure Environment: Create a .env file in the root directory:

   TWILIO_ACCOUNT_SID=your_sid
   TWILIO_AUTH_TOKEN=your_token
   ELEVENLABS_API_KEY=your_api_key
   RUNPOD_ENDPOINT=your_websocket_url

The Team

Developed for LA Hacks 2026.