-
-
Roadmap of our AI backend
Inspiration
We aim in transforming the educational system in the United States. We would like to make learning opportunities widespread. You could be a student taking daily college courses, or someone that dreams big.
What it does
Our software redefines major issues with online education. This includes a lack of engagement with instructors, less efficient learning, and weaker relations between students and instructors. By appealing to 3 main user types, our software helps administrators view teacher and student activity. Teachers will be able to gain unprecedented insight into student difficulties while students can prepare for classes with specialized information from trained AI models. Administrators are able to set up teacher and student accounts, with derived insights on the efficiency of lectures and follow-up messages. Teachers, using human-expression AI logs and smart video tracking, are able to pinpoint difficult topics for students, and AI-driven suggestions for anonymous follow-ups with students are then given. Lastly, students benefit via a 24/7 Teaching Assistant Bot, which is trained on class textbooks, professor resources, and video transcripts. This provides students with the opportunity to let their questions be heard and learn at their own pace, while not sacrificing the time of the professor and other students.
How we built it
For the frontend of the web platform, we utilized EJS, CSS, and JavaScript to deliver the user experience, which included pages for administrative tasks, navigation between courses/lectures for teachers, and a streamlined experience for students. In terms of the backend, we utilized Node.js, Express, MongoDB, and API calls managing our AI and data storage infrastructure. Several actors forced to utilize different representation for our data. Course content is stored in chroma.db, a vectorial database, in the form of coordinates in an embedded space representing the courses' material. We communicated between the user, LLM, and vectorial database through the use of AI agents. We also made sure that context was offered between each transaction with the user and database.
Challenges we ran into
We definitely had challenges integrating the backend with the AI, especially with connecting and sending data between the database and agent/LLM. Ensuring precision and performance with vector database has been quite the challenge, as degenerating in query time can be common. The frontend nearly consisted of 20 pages, making it tedious to route.
Accomplishments that we're proud of
We're proud of being able to integrate the LLM, agents, and vectorial database into a functional AI backend. We're also amazed that the majority of our frontend is built, since we were very ambitious with the amount of pages included.
What we learned
We definitely learned about integrating a lot of useful tech stacks that we haven't been exposed to
What's next for Omnia
We're definitely planning to scale up and add additional features. Such features include facial recognition to tell whether or not students are engaged.
Log in or sign up for Devpost to join the conversation.