/pdf-chat

Easily converse with your PDF documents using an LLM-powered chatbot through PDF Chat.

Primary LanguageTypeScript

Pdf Chat

Pdf chat is a chat application that lets you communicate with your PDF files using an advanced AI chatbot. This chatbot uses the RAG framework and relies on a Mongo DB database for efficient searching, along with Cohere LLMs to provide smart interactions with your PDF knowledge base.

Login Collections Collections Chat Bot Chat Bot Theme Chat Bot Theme

Features

  • Authentication: Securely log in and register.

  • Create Collections: Organize your pdfs into collections for better management and categorization.

  • Add Pdf files: Easily upload pdf files to a collection.

  • Embeddings with Cohere LLM: Utilizing the Cohere LLM (Large Language Model), the app creates embeddings for the pdf files.

  • Chat with Your Files: Engage in conversations with individual files. Ask questions or discuss about the content directly.

  • Chat with Collection: Interact with all the files within a specific collection.

  • Knowledge Base Chat: Finally all the files across all the collections.

Project Overview

  • Backend - Houses the Node.js REST backend, written in TypeScript and built with the Fastify.js framework.

  • Frontend - The UI of the app is built with React using TypeScript. Utilized the mantine.dev library for UI components, and React Query for fetching data.

  • Serverless - Contains the serverless code written in TypeScript. Utilizing AWS SAM this component hosts a step function. Triggered by events from AWS EventBridge, the step function performs tasks such as creating summaries and embeddings from the pdf file. This serverless architecture ensures for efficient asynchronous tasks.

Tech Stack

  1. Full stack Typescript:

    • React with TypeScript for a dynamic and type-safe user interface.
    • Fastify.js with TypeScript for building a scalable and maintainable Node.js backend.

  2. Event Driven Programming:

    • AWS EventBridge for seamless event-driven programming. When a user adds a record to MongoDB Atlas, an event triggers a serverless step function.

  3. Serverless:

    • AWS SAM (Serverless Application Model) with TypeScript for the serverless stack.
    • Serverless step function for handling asynchronous tasks such as creating embeddings and summarizing the pdf and saving it to MongoDB Atlas.

  4. Database:

    • MongoDB Atlas as the database platform.
    • Using MongoDB Atlas search for vector search.
    • MongoDB SDK for Node.js, offering a straightforward database interface.

  5. Cohere LLM Platform:

  • Utilizing Cohere LLM endpoints for creating embeddings.
  • Using Cohere Chat endpoint for chatbot functionality.
  • Employing Cohere Re-rank endpoint to increase the relevancy of documents fetched using MongoDB Atlas vector search.

App Workflow

  1. User Login:

    • After logging in, the user enters the dashboard view.

  2. Collection Creation:

    • In the dashboard, the user can create collections, for grouping related files.

  3. Collection Management:

    • Collections are displayed on the left menu. The user can select a collection to view and manage its files, including adding new files.

  4. Pdf file Upload:

    • When a user uploads a new pdf to a collection, the data is saved to MongoDB Atlas and it triggers an event to AWS EventBridge.

  5. Serverless Processing:

    • AWS EventBridge triggers a serverless step function, which summarizes and creates embeddings using Cohere LLM. The data is then saved to MongoDB Atlas.

  6. Chat Functionality:

    • Users can switch to the chat tab, where they can engage in conversations with all the files in a collection. The backend utilizes the RAG framework with the Cohere LLM platform for intelligent chat functionality.

  7. Pdf Detail Page:

    • Clicking on a pdf file takes the user to the detail page. Switching to the chat tab on this page allows the user to engage in real-time chat with the specific file.

  8. Global Chat:

    • On the main dashboard, users can switch to the chat tab to engage in conversations with all the files they have uploaded.

Getting Started

To run the App locally, follow these steps:

  1. Clone the Repository:

    git clone https://github.com/yaldram/pdf-chat.git

  2. Start the Frontend and Backend Servers:

    cd frontend
npm install
npm run dev
    cd backend
npm install
npm run dev

  3. Create and Configure the .env File:

    Create a .env file in the root directory of the both the folders and copy over the variables from .env.sample

  4. Serverless:

    • cd serverless/src.
    • npm install
    • cd ..
    • sam build
    • sam deploy --guided --capabilities CAPABILITY_IAM

Improvments

  • Leverage MongoDB Atlas powerful search capabilities to enhance the search functionality for both collections and individual files.
  • Implement pagination in the pdf files section to enhance the user experience.
  • Improve the chat experience by displaying references for generated chat responses. Providing context or sources for the chatbot's responses.
  • Continuing App Improvements

Challenges

  • Despite using Cohere's Chat API with conversationId to manage history, frequent hits on the token limit have been observed.
  • The current approach involves creating a new conversationId when the frontend component mounts, with conversations stored in local storage.
  • To address this challenge, exploring alternatives such as LangChain and persisting the conversations may be considered.

Contribution Guidelines

We welcome contributions from the community! If you'd like to contribute to the project, please follow these guidelines:

  • Fork the repository and create a new branch for your feature or bug fix.
  • Make your changes and submit a pull request.
  • Provide a clear and detailed description of your changes in the pull request.