MnemoMind: Chat with Your Documents & Internet

MnemoMind is a powerful demonstration of Retrieval-Augmented Generation (RAG) using the Google Gemini API's FileSearch tool. Simply upload any text-based document—from dense product manuals to lengthy reports—and start a conversation. MnemoMind intelligently searches your documents to provide precise, context-aware answers.

The application also features a seamless integration with GoogleSearch, allowing you to switch from querying your private documents to asking questions about the world, with answers grounded in up-to-date web results.

✨ Features

• Document Upload: Supports PDF, TXT, and Markdown files.
• Sample Documents: Instantly start chatting with pre-loaded examples (Hyundai i10 & LG Washer manuals).
• File-Based Chat: Leverages the FileSearch tool to perform RAG on your uploaded content.
• Web-Based Chat: Switch to GoogleSearch for real-time, web-grounded answers.
• Source Citing:
  • FileSearch: View the exact text chunks from your document that were used to generate the answer.
  • GoogleSearch: Get direct links to the web pages that sourced the information.
• Dynamic Suggestions: Automatically generates relevant example questions based on your document's content to help you get started.
• Clean & Responsive UI: A modern, intuitive chat interface built for a seamless user experience.
• Automatic Cleanup: All uploaded documents and their indexes are automatically deleted when you end the chat or close the browser tab, ensuring privacy and cost-efficiency.

🚀 Core Technologies

• AI & Backend:

  • Google Gemini API (@google/genai): The core of the application.
  • FileSearch Tool: Manages document chunking, embedding, indexing, and retrieval for the RAG pipeline.
  • GoogleSearch Tool: Provides web search capabilities for grounded, up-to-date answers.

• Frontend:

  • React: A JavaScript library for building user interfaces.
  • TypeScript: For static typing and improved developer experience.
  • Tailwind CSS: A utility-first CSS framework for rapid UI development.

⚙️ How It Works: The RAG Pipeline

MnemoMind provides a simplified and powerful RAG experience by abstracting away the complexities of the pipeline.

1. Ephemeral Index Creation: When you upload documents, the app creates a temporary FileSearchStore via the Gemini API. This store acts as a private, short-term index for your chat session.
2. Document Processing: The files are uploaded to this store. The FileSearch service automatically handles:
   • Chunking: Breaking the documents into smaller, semantically meaningful pieces.
   • Embedding: Converting each chunk into a numerical vector that captures its meaning.
   • Indexing: Storing these vectors in a way that allows for efficient similarity searching.
3. Querying: When you ask a question:
   • Your query is sent to the Gemini model with the FileSearch tool configured to use your session's store.
   • The tool performs a vector search to find the most relevant document chunks related to your question.
4. Augmented Generation: The model receives your question plus the relevant chunks as context. It uses this augmented information to generate a highly accurate and relevant answer.
5. Session Teardown: The FileSearchStore is automatically deleted when you click "New Chat" or close the browser, ensuring your data is not persisted.

📁 Project Structure

.
├── public/
│   └── index.css             # Custom CSS styles (scrollbar, settings menu)
├── src/
│   ├── components/           # Reusable React components
│   │   ├── ChatInterface.tsx   # The main chat UI
│   │   ├── UploadModal.tsx     # File upload and sample selection modal
│   │   ├── icons/              # SVG icon components
│   │   └── ...                 # Other UI components
│   ├── services/
│   │   └── geminiService.ts    # All communication with the Gemini API
│   ├── App.tsx                 # Main application component, state management
│   ├── index.tsx               # Application entry point
│   └── types.ts                # TypeScript type definitions
├── index.html                  # Main HTML file with Tailwind CSS config
├── metadata.json               # Application metadata
└── README.md                   # You are here!

⚖️ License

This project is licensed under the MIT License. See the LICENSE file for details.

Built With

• filesearchtool
• gemini
• react
• typescript