911 Coach AI

Inspiration

Emergency dispatchers are essential in critical situations, yet their training often relies on outdated methods and resources. Research shows that 25% of dispatchers feel underprepared for daily emergencies, highlighting a critical gap in training effectiveness. With the increasing complexity of emergencies, there is a pressing need for modern, adaptive training solutions to ensure dispatchers are well-equipped to handle diverse and challenging scenarios.

This project is part of Track 04 option 02.

What it does

911 Coach AI is equipped with essential features tailored for emergency medical dispatchers:

Simulation: Facilitates realistic 911 call simulations based on a diverse range of scenarios stored in the vector database. This allows dispatchers to practice and refine their response strategies in a controlled environment.
Q&A Feature: Provides immediate responses to user queries by leveraging a curated knowledge base of articles and books on medical emergencies. This ensures dispatchers have access to accurate information during critical situations.
Feedback Feature: Analyzes dispatcher conversations in real-time against established protocols and instructions from the knowledge base. By offering constructive feedback, the system helps improve dispatcher performance and adherence to best practices.

How we built it

We developed a comprehensive systems architecture design to visualize the communication flow across all components of our system (you can see the architecture diagram in the Readme file of the GitHub repository).

We developed DispatchAI using a comprehensive tech stack:

Frontend

React: Framework for building the responsive user interface.
Firebase: Provides authentication services to secure user access.

Backend

NodeJS: Runtime environment for the server-side application.
MongoDB: NoSQL database for storing user information and simulation data securely.
FastAPI: High-performance framework for building AI-driven APIs.
Pinecone: Vector database used to efficiently store and retrieve all medical scenarios, books and articles.
Cohere LLM: Utilized for embeddings and generating responses to user's query as well as generating a feedback.
AI21: Utilized to simulate a 911 call with the user.

Challenges we ran into

Acquiring relevant data and books on medical emergencies posed significant challenges.
Iterative prompt engineering was time-consuming as we optimized the use of LLMs.
Integration of AI endpoints into the web app was complex due to interdependencies between them.
Limited access to premium OpenAI keys, and latency and quality issues with free trial keys from Cohere and AI21 compared to other providers like OpenAI and Gemini.

Accomplishments that we're proud of

Successfully implemented all features including the 911 call simulator, Q&A functionality, and AI feedback to create an immersive learning experience for dispatchers.
Established a comprehensive knowledge base that underpins the simulator, Q&A, and feedback features.
Designed the application to cater to dispatchers of varying experience levels, ensuring usability from beginners to seasoned professionals, each benefiting uniquely from the app.

What we learned

Integrating multiple technologies to build a cohesive and functional system.
Harnessing the potential of LLMs to tailor specific solutions using RAG (Retrieve, Adapt, Generate).
Collaborating effectively across diverse backgrounds to develop an end-to-end functional solution.