Character Insight Extraction Using LLM (DeepStack Assignment)

Demo Video

DeepStack_Video_demo.mp4

Technologies Used

1. LangChain and Python

• LangChain:
  • Provides a robust framework for LLM integration, chaining prompts, and managing workflows..
• Python:
  • Core language used for implementing the project, ensuring flexibility and integration with ML frameworks.

2. Large Language Models (LLMs)

• Model: mixtral-8x7b (Inferenced using GroqAPI)
• Used for generating summaries, identifying relationships, and classifying character types.

3. Vector Databases

• ChromaDB: Stores and retrieves text embeddings for efficient context-based search.
• Embeddings: Generated using HuggingFace's all-MiniLM-L6-v2 model for high-quality semantic understanding.

4. Text Splitting

• RecursiveCharacterTextSplitter: Splits large text into manageable chunks with overlaps to ensure continuity of context.

---

File Structure

Installation

1. Clone the repository:

   git clone https://github.com/VivekShinde7/Character_Insight_Extractor_Using_LLM.git

   cd Character_Insight_Extractor_Using_LLM

2. Set Up a Python Environment:

    conda create -prefix ./env python=3.9 -y

   conda activate ./env

3. Install Dependencies:

    pip install -r requirements.txt

4. Set Up Environment Variables:

   • Create a .env file in the project root:

    HF_TOKEN = your_huggingface_token
    GROQ_API_KEY = your_groq_api_key

Usage

1. Preprocess the Text

• Use the compute_embeddings.py script to process your text and store embeddings in the vector database:

  python src/compute_embeddings.py data/

2. Analyze a Character

• Run the get_character_info.py script to extract character details:

  python src/get_character_info.py "<character_name>"

• Example

  python src/get_character_info.py "Eliza"

• Output

Future Enhancements

• Integration with Neo4j: Enable graph-based relationship visualization to provide a clearer and more interactive representation of character connections.
• Fine-Tuning the LLM: Improve role classification and relationship detection accuracy by fine-tuning the LLM on a domain-specific dataset.