Pastelize (WIP)

RAG-based 2nd Brain Application

TL;DR

Pastelize is a RAG-based 2nd Brain application that allows users to create a personal knowledge base by creating mindmaps. All mindmaps are stored in a local Vector Database and are queried automatically when users edit mindmaps.

The part on the right side of the photo shows the results of searching for similar contexts from all vectors stored based on the current work content.



PDF Viewer

Users can upload PDF files and view them in the application. All uploaded PDF files are processed by local Embedding Model and stored in a local Vector Database as vectors. (vectors would be used for RAG, and similarity search)

When a PDF fragment is found through similarity search, users can easily navigate to the corresponding location to see the original content.



Technical Details

  • Langchain
  • Hugging Face
  • Typescript
  • React
  • Electron
  • Tailwind CSS



Technical Challenges

  • Mindmap Performance: It was important to make mindmap editing smooth and responsive.
    • For better performance, we normalized the redux states.
  • Embedding Model is quite heavy: We used a local Embedding Model to process PDF files and extract text, images, and other content for embedding and storage.
    • When the electron app calls the embedding model, it was really heavy and made the app freeze for a while.
    • To avoid this, we used Web Worker to run the embedding model in the background.
  • Seamless UX: It was hard to move away from development-level UI/UX details.
    • We tried to make the UX seamless by adjusting the animation and transition details.