EmbrAlrt

• 

  Mobile App View

• 

  Main Homescreen

• 

  Upload Images to detect wildfire using our lightweight RNN

• 

  Live Wildfire Analysis

• 

  RAG chatbot with Wildfire safety + Santa Clara County Specific information

Inspiration

We chose the Public Safety track as we wanted to apply our skills in creating a project that could help protect and support the broader community. EmbrAlert was inspired by the devastating LA fires earlier this year, which personally impacted many of our family and friends. Witnessing how quickly wildfires can spread, and how devastating the consequences were, we researched San Jose’s history and discovered that Santa Clara County has lost over 1,500 buildings to wildfires in just the past five years. This made it clear that we needed to build a solution that was both reactive and preventative to better safeguard our communities.

What it does

EmbrAlert is a one-stop platform for wildfire detection, prevention, and community alerts, built for the diverse San Jose area and beyond. It offers real-time wildfire risk assessments, live weather and air quality updates, and a simple dashboard for users to interact with. Users can upload images of potential smoke, and our lightweight RNN model predicts wildfire likelihood. The app also features a multilingual chat tool powered by a custom RAG pipeline, supporting six languages common in San Jose, with both voice and text input. Live camera wildfire detection without uploads is also integrated for instant reporting.

How we built it

We built the frontend using React, React Native, and Vite, while the backend relies on Python, LangChain, and LangGraph. The RAG pipeline connects to a vector database for intelligent information retrieval. Our image analysis is driven by a conventional RNN model, and live camera/voice functionalities are tightly integrated into the app experience.

Challenges we ran into

Our biggest challenges were building a real-time multilingual chat system, changing the UI to handle live interactions smoothly, integrating the camera with API calls, and connecting the vector database with the RAG pipeline in LangChain to interface with the LLM. Balancing responsiveness while maintaining deep, accurate information retrieval was a key technical hurdle.

Accomplishments that we're proud of

We successfully created an accessible wildfire detection tool that works live across languages, devices, and input types. We packaged AI-driven wildfire prevention education, community alerts, and live smoke detection into a platform that's easy for anyone to use — no tech expertise required.

What we learned

We deepened our skills in building scalable RAG pipelines, integrating vector databases with LLMs, optimizing React Native apps for performance, implementing live camera and voice features, and designing multilingual AI systems. Working with real-time hardware integration for computer vision also gave us valuable hands-on experience.

What's next for EmbrAlert

Next, we plan to launch EmbrAlert to the public, integrate emergency authority notifications, enhance our RNN model with additional wildfire datasets, and expand language support. Our goal is to make EmbrAlert a scalable solution for wildfire-prone regions around the world.

Built With

• cag
• langchain
• llm
• python
• rag
• react
• vite