Cognimate - Team 52 [5E]

Cognimate
Outcome of course builder
Outcome of Summariser

About us

We are a group of students from diverse backgrounds (Computer Science, Math, IEM) working on the project together. We are Team 52 Reference Code 5E, and we have built our very own GenAI platform Cognimate.

Inspiration

Inspired by the idea of self-driven learning, we aim to provide accessible education to students in Singapore irregardless of their background and disability to provide students with equal opportunities in learning.

We noticed that access to educational material is not equal to all students in the age of subscription-based business model such as Chegg, coupled with the high cost of private tuition fees, educational materials are increasing harder to access in Singapore. We aim to design a system which provides driven students an equal footing in access to study materials that they might otherwise not have.

What it does

A Generative AI (GenAI) Course Builder and Summariser that allows students to curate personalised educational material for self-study and revision purposes. We provide 2 main functionality (1) Course building and (2) Course summarising. In our course builder, students can define the course and sub-topics they want to build, Cognimate will aid students in generating chapters and source for top videos that students can browse to learn more about the topic. As for the course summarisation, students will upload a PDF which will be summarised by Cognimate. To further value add, quiz questions will be provided and a chatbot preset with context is made available to enhance learning.

How we built it

Our tech stack includes the use of popular tools such as Next.js and tRPC which is a safer implementation to reduce the likelihood of bugs in our program. gen ai architecture

On GenAI, we utilised APIs from OpenAI to generate textual response and Unsplash for thumbnails. We designed Cognimate as a AI driven platform, whereby the content of our platform is highly personalised to the profile of the student. We chose a purely GenAI approach to reduce the barrier to education where students can freely print webpages into PDF to generate courses for their own learning.

gen ai diagram

Challenges we ran into

We were not able to get the generative models to output consistent questions and answers that conform to Singapore's educational system methodology. We had to perform different types of prompt engineering and tweaking of our solution to collect the required data which eventually acted as parameters for giving a meaningful prompt to generate personalised content for students.

Accomplishments that we're proud of

Completing the hackathon in a short 2 days is no easy feat for us and others. We are really glad that we came together as a well-diversified team to tackle this problem statement from various angles and eventually coming up with a holistic solution.

What we learned

We learnt how to design a GenAI driven platform which is vastly different from the traditional rule-based programming and routing. It was also very interesting to continuously tune our prompts into the Generative models to ensure a consistent and reliable output. Beyond technical capabilities, we also had fun learning other tools such as Figma and video editing for our youtube demo.

What's next for Cognimate

In order to reduce cost overhead and load-time, we can shift our application to use open source models such as Llama which are deployed on self-hosted GPU clusters, allowing for more concurrent users and scalability.
Extending the use case from Singaporean students to other countries. This would include provision of support for languages other than English and to also deep dive into the needs of students for different counties.