ReelKnowledge

💡 Inspiration

Short-form platforms like Instagram Reels, TikTok, and YouTube Shorts contain massive amounts of valuable knowledge—tutorials, career advice, coding walkthroughs, business insights, science explainers, etc.

But for users who want to learn, not just scroll, these platforms are deeply inefficient:

Videos are unsearchable once watched
Meaningful information is buried inside long audio
Saving and organizing content is tedious
Rewatching is required to extract key points
There’s no way to build a “knowledge base” from what you consume

As a result, millions of people consume high-value content every day but retain almost none of it. This is the learning-efficiency gap we set out to solve.

🚀 What it does

ReelKnowledge is an app that solves this problem. A user shares a video link (from TikTok, Reels, etc.), and our backend pipeline gets to work. It extracts the audio, transcribes it, and then uses the Claude-Sonnet 4.5 API to create a concise summary and a list of key takeaways. This summary is then sent to the user's app, creating a clean, organized, and searchable library of the knowledge they actually wanted to save.

🔧 How we built it

We built the project with a monorepo structure, split between a Python backend and a React Native frontend.

Backend: We used FastAPI as our web server. It runs a pipeline that extracts audio (using ffmpeg), transcribes it, and then sends the transcript to the Claude-Sonnet 4.5 API for summarization. We used Supabase as our database to store the user data and summaries.
Frontend: The mobile app was built using React Native (Expo) with TypeScript. We used NativeWind (Tailwind for React Native) to style all of our components, including the Login Page, Home Page, and the Video Summary Modal.

🤯 Challenges we ran into

We tried to deploy our backend on Heroku and spent over an hour on this. Sadly, we found that basically every social media website was blocking requests from the Heroku app due to their bot protections. We tried sending cookies for a logged-in session, but this still didn’t work. Given our time constraints, we opted for hosting the backend locally instead.

🏆 Accomplishments that we're proud of

We are proud of successfully building a full-stack, AI-powered mobile application within the time constraints. We created a complete data pipeline that connects a React Native frontend, a FastAPI backend, a Supabase database, and the Claude API. We also built a clean, multi-page mobile UI with a working login, home dashboard, and a video summary modal.

📚 What we learned

Our main takeaway was the difficulty of deploying backends that interact with major social media sites. We learned that platforms like Heroku are often on blocklists due to bot protections, making scraping or data extraction impossible. This forced us to be flexible and pivot to a local-first hosting solution to prove our concept. We also learned how to effectively divide work for a complex full-stack project among team members.

⏩ What's next for ReelKnowledge

With more time, we wanted to evolve the app into a full AI-powered personal knowledge engine. Our biggest planned feature was a RAG (Retrieval-Augmented Generation) system. This would let users ask any question and receive answers grounded in the videos they’ve previously summarized. By embedding the user’s query and searching across their stored video notes, the app could act as a personalized “second brain,” making all of a user’s consumed content searchable and actionable.