BookMind

BookMind is a web application that allows users to explore character relationships and storylines in books using AI-powered visualizations. The application provides interactive mind maps, AI chatbots for deep questions, book summaries, and community contributions.

Features

• Interactive Mind Maps: Visualize relationships between characters and plot elements.
• AI Chatbot: Ask deep questions about the book and get insightful answers.
• Book Summaries: Get concise overviews of plots and themes.
• Community Contributions: Add and refine maps with fellow book lovers.

Prerequisites

• Node.js
• Python >= 3.10
• LlamaStack server running locally
• Environment variables:
  • LLAMA_STACK_PORT
  • INFERENCE_MODEL
  • REACT_APP_GOOGLE_BOOKS_API_KEY

Getting Started

Backend Setup

1. Install dependencies:

pip install -r server/requirements.txt

2. Install and run our forked llama-stack docker

git clone https://github.com/seyeong-han/llama-stack.git
cd llama-stack
pip install -e .

# Get your API_KEY from https://cloud.sambanova.ai/apis
export SSAMBANOVA_API_KEY=YOUR_API

llama stack run --env INFERENCE_MODEL=meta-llama/Llama-3.1-8B-Instruct sambanova

3. Run the server:

python server/server.py

Frontend Setup

1. Install dependencies:

npm install

2. Run the application:

npm start

Usage

1. Initialize Memory: Upload your book or choose from the library to initialize memory.
2. AI Analysis: The AI analyzes the book and generates a mind map.
3. Explore Insights: Explore relationships, themes, and Q&A insights.

What did we use Llama-stack in BookMind?

1️⃣ Implemented SambaNova Systems Inference API: We contributed to the open-source Llama-stack by integrating the inference API and built a custom Docker container for the llama3.2-3B-instruct model.
2️⃣ RAG with FAISS: We leveraged FAISS in Llama-stack for Retrieval-Augmented Generation, enabling real-time responses to character relationship questions.
3️⃣ Multi-Hop Reasoning: Our system performs sequential inference—first extracting characters and relationships, then generating graphized mind map data in JSON for visual storytelling.