wangyixing

HOME
agent page
Q&A

💡 Inspiration

The inspiration for sophie Learning Agent comes from a pain point shared by lifelong learners everywhere: The fragmentation of knowledge.

We realized that while learning happens everywhere—on paper notebooks, in PDF slides, and on whiteboards—our digital brains (like Notion) are often disconnected from these physical sources. Students and professionals waste countless hours manually transcribing text and organizing notes, which interrupts the "flow" of learning. We wanted to build a bridge that turns the physical world into a structured digital knowledge base instantly.

💻 How we built it

We built sophie using a modern, decoupled architecture centered around the Model Context Protocol (MCP).

The Core: We utilized LangGraph to orchestrate an intelligent workflow that treats different tools as modular nodes.
The Eyes: We integrated PaddleOCR (via Python) to handle high-precision text and table recognition from images.
The Brain: A Node.js backend manages the logic, processing the OCR data and structuring it.
The Hands: Through the Notion API, the agent automatically creates standardized, tagged, and formatted notes in the user's workspace.
The Face: A React + Vite + TailwindCSS frontend provides a clean, responsive interface for users to upload materials and monitor the agent's progress in real-time.

🚧 Challenges we faced

The biggest challenge was time. We discovered the hackathon late and had only 72 hours to build this from scratch.

Integration Complexity: Making a local Python environment (for PaddleOCR) talk smoothly with a Node.js MCP server required precise configuration and error handling.
Quality Control: With such a tight deadline, my teammate @Sophie618 and I had to manually review almost every line of AI-assisted code to ensure the logic was sound and the product was stable.
Workflow Orchestration: Designing a LangGraph flow that could handle failures gracefully (e.g., if OCR misses text) was a steep learning curve.

📚 What we learned

This intense 3-day sprint taught us the true power of Agentic Workflows. We learned that the future of AI isn't just about chatbots, but about interconnectivity. By adopting the MCP standard, we realized we weren't just building a tool, but an ecosystem where any OCR engine or Knowledge Base could be swapped in.

We also learned the importance of focusing on the "last mile" of user experience—ensuring that the transition from a raw image to a beautiful Notion page feels like magic.

We are proud to present a working MVP that demonstrates the potential of automated knowledge management.