Inspiration
This is a huge problem that the three of us encountered while studying for our CMPSC 497: Deep Learning for Computer Vision Final Exam last spring. We had 4500 slides to go through and weren't really sure what to focus on. Luckily with one day to go, a friend of ours who took the course last year sent us their notes that they made specifically for the final exam had everything we needed to get through it, and we want Memoire to be that friend for everyone else!
What it does
The platform goes through the uploaded PDFs and then makes concise notes based on the important topics present in the chapters. It also has a learn mode which also helps the students learn the concepts and topics and can even test their understanding with a quick quiz.
How we built it
From conceptualization to implementation, this project was a dynamic journey. We began by whiteboarding our systems design, laying the foundation for a robust architecture. Our user interface was meticulously crafted in Figma, aligning with the ShadCN component library to ensure consistency and efficiency in our design process. The core of our product was built on a modern tech stack, leveraging React for interactive UI components, Next.js for server-side rendering and routing, and TypeScript for enhanced code reliability and maintainability. To power our advanced functionality, we integrated LLaMA 3 (Large Language Model Meta AI) through LangChain, enabling sophisticated natural language processing capabilities. Finally, we utilized Flask as our web framework to seamlessly integrate these components, creating a cohesive and powerful application. This comprehensive approach allowed us to navigate the challenges and deliver a cutting-edge product that met our ambitious goals.
Challenges we ran into
Throughout the development process, we encountered significant challenges with our core functionality, particularly in terms of performance optimization. To address these issues, we engineered an efficient pipeline and streamlined our workflow, resulting in substantial improvements in overall system performance.
Managing dependencies proved to be a complex task, as they were initially scattered across various components of our project. The process of identifying, organizing, and properly configuring these dependencies was time-consuming but ultimately crucial for ensuring a stable and maintainable codebase.
Eggs tried to murder Chinmay
Accomplishments that we're proud of
Brought our pipeline runtime down from 20 minutes to 2 minutes. We built a sleek UI and were able to translate that into code. We achieved a combined sleep time of 10 hours, officially making us Zombies.
What we learned
We learned how to work with LangChain more specifically, how to prompt engineer it to provide its response in a format ideal to export it into our DB for our requirements.
What's next for Memoire
We want Memoire to become the one-stop-shop for all students while taking course and especially when studying for a big exam. Gone are the days where you spend minutes looking through 10 slides to find one valuable point, Memoire will summarize and highlight the key points as well as help you learn and understand them. Your memory might fail you, but Memoire won't!
Log in or sign up for Devpost to join the conversation.