PrepPal

Inspiration

The idea for PrepPal was born out of the increasing frequency and severity of natural disasters that disrupt lives and leave communities in desperate need of essentials. We wanted to create a tool that leverages AI to give people rapid access to life-saving resources like water, food, and shelter information in these critical moments. Our interdisciplinary team saw an opportunity to bring together technology, emergency preparedness, and user-centered design to help individuals and communities prepare for, respond to, and recover from disasters with greater resilience.

What it does

PrepPal is an AI-powered companion designed to support communities before, during, and after natural disasters. The app provides personalized disaster-specific checklists, real-time resource mapping, and essential guidance based on individual needs. Powered by IBM's watsonx.ai platform, PrepPal gives users timely and relevant information and connects them to resources like nearby shelters, food supplies, and first aid centers—all accessible on their phones. Beyond real-time aid, PrepPal has educational potential, helping users and organizations improve disaster preparedness and resilience through real-world data and insights.

How we built it

Our tech stack includes React for a responsive frontend experience and FastAPI in Python for backend processes, ensuring smooth performance. We used IBM’s watsonx.ai platform, particularly the Granite LLM, and an embedding model called Slate, which enables PrepPal to retrieve context-rich, accurate information tailored to specific disaster scenarios. We implemented a Retrieval-Augmented Generation (RAG) workflow, creating vector embeddings from publicly available disaster preparedness PDFs and storing them in PineconeDB. This RAG setup allows PrepPal to retrieve and incorporate relevant data from these documents into real-time user predictions, enhancing accuracy and context for critical responses. Challenges we ran into

One challenge was optimizing the RAG workflow to effectively pull relevant data in real-time without latency, ensuring that users receive information quickly during emergencies. Integrating the IBM Slate model with PineconeDB and IBM’s Granite LLM also required careful tuning to maintain accuracy across a variety of disaster contexts. Another challenge was the interdisciplinary collaboration, balancing our technical development with business strategies to ensure the app would truly meet user needs in crisis situations.

Accomplishments that we're proud of

We’re proud of our team’s ability to integrate AI into a practical, user-centered solution that could make a significant impact in emergency preparedness and response. Our RAG workflow with PineconeDB and IBM's Granite LLM setup enhances PrepPal’s ability to provide relevant information with high accuracy. We also successfully collaborated across multiple disciplines, merging technology, business insights, and design to create a solution that’s both impactful and adaptable to various disaster scenarios. What we learned

Throughout this project, we gained valuable experience in implementing RAG workflows to enhance language model accuracy in real-time applications. Our team deepened its knowledge of disaster management needs and the importance of accessible, personalized resource information. Collaborating across technical and non-technical fields also reinforced the value of interdisciplinary perspectives in designing solutions that are technically sound and user-friendly.

What's next for PrepPal

Moving forward, we plan to expand PrepPal’s capabilities by integrating additional APIs for location-based resources and developing predictive analytics that offer early warnings based on environmental and historical data. We also aim to work with emergency management professionals to refine our guidance and add multilingual support for broader accessibility. With continued development, PrepPal has the potential to not only assist in immediate disaster response but also contribute to long-term community resilience through preparedness education.

Built With

Submitted to

UF/IBM AI Days Hackathon Supported by Swamphacks! https://discord.gg/4EJhabM79p

Created by

I collaboratively built the full-stack infrastructure to deliver disaster preparedness guides to users. I configured the vector database to handle queries and store unique records, contributed to developing the user interface, and designed the data structures to efficiently communicate with the backend.

Max Martin
I worked on parsing PDF files, contributed to the frontend, and (mostly) deployed the website at http://52.116.197.232/

Tuan Dang
Full stack development, Google Maps API queries and integration for location-searching, frontend design and components

Zachary Tenn Yuk
CS at UF
I enjoyed learning about Watson, contributing to the backend server that implements a RAG workflow. I created scripts to process Disaster documents, embedding their content using IBM Watson's Slate model and storing these embeddings a vector DB. To integrate the client, I embed the incoming query with Slate, retrieve relevant contextual from vector DB, and generate augmented responses using Watson's Granite model.

John Midel Miclat
I was responsible for overseeing the creative direction, design, and business aspects of the project. This involved ensuring a strategic integration of technical components with the needs of end users, focusing on delivering a user-friendly experience.

Elizabeth de Jesus