CalmCampus AI

Inspiration

We are students who have now transitioned into the phase of being parents with young children. Our experiences from both stages in life—learning in academic environments and now ensuring the safety of our children—have made us more conscious of the importance of preparedness, especially in emergency situations. Schools and colleges are places where students should feel safe, but we know that emergencies can happen unexpectedly. Our desire to create a solution that helps parents, students, and administrators be more informed and confident in such situations is what inspired us to build CalmCampus AI.

What it does

CalmCampus is designed to reduce stress and anxiety regarding emergency situations. It brings a sense of peace and confidence by providing clear, actionable information on how to respond when emergencies occur. The platform allows users to quickly search for information and find out if their loved ones are safe or accounted for, significantly reducing uncertainty and panic. And it works no matter which language the user speaks or writes.

How we built it

We built CalmCampus using a combination of external libraries, OpenAI’s natural language models, and custom code written in PHP and JavaScript. By integrating these technologies, we were able to create an interactive user experience that serves both regular users (like students & parents) and site administrators. The system handles real-time responses and manages queries by interacting with a local database, reducing dependency on external APIs to improve performance.

How it works

First, school administrators will upload their school emergency response plans to their OpenAI v2 Assistant. Currently, this is done by logging into the OpenAI console; however, in future versions, an upload option can be added to the CalmCampus Dashboard for easier access. Once they've uploaded their content, they can go to their CalmCampus Dashboard and start manually adding questions with answers or generating questions. If having the system generate the questions/answers, questions are generated by having the Assistant study the content of the emergency response plans and identifying the most common and important questions to have responses for. After questions are generated, the administrator has the ability to select the questions and receive AI-generated responses. The generated responses come in two forms: 1) Plain text, which is used for the chatbot, and 2) Rich text, which is used to format the page.

The reason we pre-generate the responses is twofold: 1) Speed of response – by storing responses in a local database, rather than requesting real-time answers from OpenAI/Gemini, we dramatically reduce response times, carbon footprint, and the underlying costs of running the AI engine. 2) Accuracy – this allows school administrators to ensure that the information being shared with students and parents is as accurate as possible. AI engines are powerful, but sometimes they don’t interpret the underlying content perfectly enough to represent something as important as a school emergency response.

When regular users (students, parents, & teachers) use the app on the front end, they are presented with both rich text (HTML) and a chatbot. The questions asked in the chatbot transform the rich text displayed above. This allows users to view important information in a user-friendly format. Whenever a user asks a question, the system first checks the local database for an exact match. If no exact match is found, it queries the OpenAI ChatCompletions engine for a similar match. If no similar match is found, it requests an answer from the OpenAI Assistant, which first generates a plain text response followed by a rich text response. Since this process can take up to 20-30 seconds, it’s recommended that school administrators pre-generate responses to the most common and basic questions.

Challenges we ran into

One of the biggest challenges we faced was optimizing the performance of OpenAI’s models. Initially, the bot was taking around 30 seconds to generate responses, which was too slow for an emergency-based application. To address this, we implemented functions that would pull common answers directly from a local database, dramatically reducing response time while maintaining accuracy.

Accomplishments that we're proud of

We are incredibly proud of coming together and actually getting this project built! It was an idea we had talked about for a long time, and turning it into a reality felt like a huge accomplishment. Each of us brought our unique experiences and skills to the table, and the fact that we are a group of individuals of faith makes this project even more special. Our shared purpose and collaboration truly powered this achievement.

What we learned

We discovered a wealth of government-published information on how to prepare for and respond to active emergencies, especially in educational settings like schools and colleges. This information was critical in shaping the content and response logic of CalmCampus, ensuring that the advice we provide is accurate and reliable.

What's next for CalmCampus AI

Our next step is to take CalmCampus AI to the market. We plan to approach local schools in Los Angeles and Atlanta to see if they would be interested in adopting this technology to better inform both students and parents during emergencies. By bringing this solution to schools, we aim to significantly reduce stress and improve communication during critical situations.