PDFpedia
Open-source chat with documents using Retrieval Augmented Generation (RAG) and powerful open-source LLMs
Timeline
2024
Role
Full Stack
Team
Solo
Status
CompletedTechnology Stack
Key Challenges
- RAG Implementation
- Local LLM Integration
- Browser-based Embeddings
Key Learnings
- Retrieval Augmented Generation
- Vector Stores
- Ollama Integration
PDFpedia: Chat with Documents
Overview
PDF-pedia is an exciting open-source chat with documents project using Retrieval Augmented Generation (RAG) implementation. It leverages powerful open-source Large Language Models and a modern tech stack to provide a user-friendly application for interacting with PDF documents.
What is PDFpedia?
PDF-pedia simplifies the way you interact with PDFs. It uses open-source Large Language Models and can run entirely on your local machine. PDF-pedia harnesses the power of an open-source large language model—it can run entirely locally, making it a privacy-focused and efficient solution for PDF document management.
Key Features
- Open Source LLMs: PDF-pedia leverages open-source Large Language Models, ensuring accessibility and transparency
- Local Execution: Run PDF-pedia on your own machine, putting you in control of your data and privacy
- Simplified PDF Interaction: Chat with your documents, extract information, and generate answers as if you're having a conversation
- Offline Support: Go offline after the site loads—all processing happens client-side
Tech Stack
- Voy: Vector store for storing and managing vectors, fully WebAssembly (WASM) compatible for browser execution
- Ollama: Runs Language Models locally and exposes them to the web app
- LangChain.js: Orchestrates all pieces—calling models, performing retrieval, ensuring smooth operation
- Transformers.js: Runs embeddings in the browser for accurate PDF processing
- Next.js: Application framework
How It Works
- Upload PDF documents to the application
- Content is extracted and processed client-side
- Embeddings are generated using Transformers.js in the browser
- Voy stores vectors for efficient retrieval
- Ollama provides the local LLM for generating responses
- LangChain.js orchestrates the RAG pipeline
Why I Built This
I built PDFpedia to create a privacy-focused alternative to cloud-based document chat tools. By running entirely locally, users can interact with sensitive documents without sending data to external servers. The combination of modern web technologies and open-source LLMs makes powerful document AI accessible to everyone.
Credits
- @dawchihliou for Voy
- @jmorgan and @mchiang0610 for Ollama
- @xenovacom for Transformers.js