Relearn

Discord contact

Nand Vinchhi - vinchhi
Advay Choudhury - pdy.gn
Rushil Madhu - rushil695

Inspiration

Over the past few years, social media platforms like Tiktok and Instagram reels have used short-form content to take over the entertainment industry. Though this hyper-optimized form of content is engaging, it has led to students having shortened attention spans and a higher tendency to multi-task. Through Relearn, we aim to modernize studying to match these changing mental abilities.

What it does

Scrolling interface for short videos

Intuitive tiktok-like interface for scrolling educational short videos
Complete with engaging music, bold subtitles, and background videos
Functionality to like videos, indicating that the user has understood that particular concept

Content generation AI

Uses lecture videos from Khanacademy + GPT-4.5 analysis to generate bite-sized educational clips
Custom video processing for adding elements such as music and subtitles

Recommendation algorithm

Ensures that the user progressively covers all content in a given topic
Occasional meme videos for a quick study break!

Edison AI

An LLM-based bot account that answers student questions in the comments section
Understands the context from the current video to provide precise and useful answers

Additional details for sponsor prizes

Best Use of AI in Education

Relearn explores the confluence of cutting-edge AI tech and useful education solutions -

Our content generation pipeline utilizes LLMs (GPT-4.5) to understand context in Khanacademy lecture videos to create short and engaging clips
Our AI account, Edison, utilizes LLMs to provide precise and useful answers to questions in the comments of a specific video

Best Accessibility Hack sponsored by Fidelity

Our platform makes high-quality content available over the internet. We enable anyone with a mobile phone to improve their study consistency and quality, which would help bridge economic and social gaps in the long term.

Most Creative Use of Redis Cloud

We used Redis for the following pivotal aspects of our app -

High speed caching and access of dynamic data such as likes, comments, and progress
Caching the latest video that a given user interacted with in, and feeding that into our recommendation algorithm (speeds up the algorithm by a huge margin)
See this file for our Redis functions - https://github.com/NandVinchhi/relearn-hackathon/blob/main/backend/dbfunctions.py

How we built it

Front-end mobile application

Built with Swift and Swift UI
Used multiple packages for specific functionalities such as Google Auth, video playback etc.

Content generation pipeline

YT Transcript API + PyTube for scraping Khanacademy videos
Open AI to retrieve context and divide long lecture videos into bite-sized clips
MoviePy for automatic video editing - subtitles, music, and background video

Back-end

Python + FastAPI for the server
Supabase database + cloud buckets for storing short videos + curriculum
Redis database for storing dynamic like, comment, and progress data
Server deployment on AWS EC2

Recommendation algorithm

Caching latest 'liked' short videos for each topic and unit in Redis
Super fast progress retrieval from in order to compute the next video on scroll

Edison AI

Used GPT 4.5 + short video transcripts for context-specific answers
AI chat endpoints deployed along with server on AWS

Challenges we ran into

LLM hallucinations
Bugs with automated video processing
Getting the video scrolling interface to work smoothly without any sort of flickering or unwanted behavior
Getting the whole stack working together and deployed before submission deadline

Accomplishments that we're proud of

Implementing a custom video scrolling interface for our short videos
Automatically generating high-quality short videos
Designing a clean user interface using Figma before building it out
Getting back-end integrated with google auth + multiple databases + APIs and deploying to cloud
All of this in < 36 hours!

What we learned

Using Redis to enhance a more conventional tech stack
Deploying FastAPI to AWS EC2 using Gunicorn workers
Building complex automated video editing functionality using MoviePy
Utilizing the newly released GPT-4.5 API to perform context retrieval and create high-quality videos.
Building custom UI Components in Swift and working with Google Auth in an iOS application