DarsyAI

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned## 💡 Inspiration

Traditional education, especially with dense textbooks like the Egyptian "El Moasser", often leaves students unengaged and overwhelmed. We realized that learning shouldn't be a one-way street. Inspired by the potential of Voice AI, we wanted to transform static, heavy curriculums into interactive, living "Study Labs" where a student can converse with their books and listen to tailored audio masterclasses.

⚙️ What it does

LevelUp (Darsy AI) is a next-generation smart private tutor. When a student uploads a textbook (even scanned images or complex Arabic PDFs), the platform instantly generates a dedicated "Study Lab". Inside this lab, students can:

• Listen to a generated Audio Masterclass (podcast-style summary) of the chapter.
• Chat in real-time with an AI tutor that strictly references the uploaded curriculum.
• Get instant summaries and quick quizzes without leaving the UI.

🛠️ How we built it

We architected a resilient, hybrid system:

• Frontend: Built with Next.js and TailwindCSS, featuring a highly polished Glassmorphism UI and a pulsing "Brain Avatar" to visualize AI processing states.
• Backend Intelligence: We completely bypassed traditional, error-prone local PDF extraction. Instead, we built a custom Node.js bridge using notebooklm-py to pipe documents directly into Google NotebookLM.
• The AI Engine: The core conversational agent is powered by Gemini 3.1 Flash Lite, ensuring lightning-fast responses, while NotebookLM handles the heavy RAG (Retrieval-Augmented Generation) and audio generation.
• Database: We used Supabase (PostgreSQL) to store user profiles and map unique notebook_ids to the students' specific study materials.

🚧 Challenges we ran into

Handling complex Arabic scanned PDFs was our biggest hurdle; local text extraction kept failing. We solved this by pivoting to a direct NotebookLM backend upload strategy, leveraging Google's superior OCR. Additionally, we faced ETIMEDOUT errors from the CLI. We engineered a robust fallback chain: if NotebookLM delays beyond 8 seconds, the system seamlessly routes the query to Gemini 3.1 directly, ensuring 100% uptime and a smooth UX.

🏆 Accomplishments that we're proud of

We successfully "White-labeled" NotebookLM's raw power into a beautifully branded, student-friendly interface ("LevelUp Private Tutor"). We managed to integrate two powerful Google AI ecosystems (NotebookLM for RAG/Audio + Gemini API for conversational speed) into a single, cohesive full-stack application.

📚 What we learned

We mastered building complex fallback logic, syncing Next.js server-side operations with Python-based CLI tools, and managing dynamic Supabase schema migrations on the fly.

🚀 What's next for LevelUp

We plan to introduce Gamification (XP, daily study missions, leaderboards) and expand the platform to support collaborative study rooms where multiple students can interact with the same AI tutor simultaneously.

What's next for DarsyAI

Built With

• google-gemini
• next.js
• node.js
• notebooklm
• react
• supabase
• tailwind-css
• typescript