adept

Inspiration

Adept was created from the frustration of navigating an overwhelming number of online learning resources. We realized that while quality content exists, finding the right guide for your unique needs is both challenging and often expensive. Adept was inspired by the vision that a personal coach who can provide tailored, actionable insights can help you master any skill.

What it does

Adept acts as your personal tutor and coach by fusing curated high-quality instructional content with advanced visual analysis. It scours trusted databases and YouTube archives to build a knowledge graph using Video RAG technology, then matches your performance against expert examples to provide personalized guidance and recommendations.

How we built it

Frontend: Developed using Python and Streamlit for an intuitive, interactive user interface.
Content Aggregation: Leveraged yt_dlp, BERT (Transformer), and Requests to source and download instructional videos.
Knowledge Graph & Analysis: Integrated cutting-edge Video RAG to extract visual context from our curated HowTo100M database and external resources.
Core Logic: Combined these components to analyze user performance in real time and offer actionable feedback.

Challenges we ran into

We had to rework a state-of-the-art VideoRAG system to work with Gemini instead of OpenAI so we could develop our project for free.
We had to work with GPUs to efficiently use the VideoRAG indexing model, which encountered numerous GPU compatibility issues due to its high computational requirements.
We initially relied on the HowTo100M dataset, which wasn't as up-to-date as we needed. As a result, we transitioned to using the entirety of YouTube as our database, necessitating the development of more efficient search algorithms to accurately retrieve the most relevant videos.

Accomplishments that we're proud of

We were the first to utilize the 3-week-old VideoRAG technology in an educational pipeline that grounds feedback in real-world
We created efficient algorithms to search through YouTube's vast repositories and find videos that are genuinely helpful for our users.
We built an intuitive interface that simulates the benefits of having a personal tutor or coach who can observe you.
We explored the use of Video RAG in an educational context, guarding our feedback against hallucinations while providing new levels of personalized depth and accuracy.
Creating technology that democratizes education and provides personalized, tailor-made guidance to underprivileged people that would be otherwise inaccessible.

What we learned

The importance of designing with the user in mind to create engaging, effective learning tools.
How to integrate diverse technologies to solve complex problems, and how to navigate the challenges that arise when combining these complex systems.
Techniques for grounding large language models in verified, quality content to minimize misinformation and improve reliability.

What's next for Adept

Content Expansion: Broaden our curated content database to include even more diverse topics and resources.
Real-Time Processing: Enable live interaction with Adept by streaming data and receiving constant responses.
Scalability: Transition to cloud deployment to offer a scalable, responsive service accessible to learners worldwide. -User Experience: Improve the user experience by allowing users to maintain and reuse separate chats