Sheldon Cooper Conversational Bot

This project lets you chat with a simulated Sheldon Cooper from the TV show 'The Big Bang Theory'. It uses a large language model (Gemini Pro) to generate responses in Sheldon's style, based on a provided context. The context is retrieved from a vector database populated with data sourced from a Kaggle dataset (https://www.kaggle.com/datasets/mitramir5/the-big-bang-theory-series-transcript), providing information relevant to the conversation. The project leverages libraries like Pathway, LlamaIndex, and Google Generative AI for data processing, embedding, retrieval, and response generation.

Link to Video Demo

https://www.loom.com/share/262ded553a2b448d8e525c22ca64fd9d?sid=b256fa82-cd45-4b63-86d5-c75aa36d2580

Libraries Used

1. Llama Index: * Node Parsing: TokenTextSplitter from llama_index.core.node_parser is used to split the data into smaller chunks (nodes) for efficient processing and embedding.

   • Retrieval: PathwayRetriever from llama_index.retrievers.pathway is used to fetch relevant context from the vector database built using Pathway. This context is then used to generate Sheldon-like responses.

2. Pathway: * Data Ingestion: pw.io.fs.read is used to read data from the file system. * Vector Database: Pathway's VectorStoreServer is used to create and manage the vector database where the embedded data is stored. This allows for efficient similarity search to retrieve relevant context.

3. Google Generative AI: * Response Generation: The gemini-pro model from google.generativeai is used to generate responses in Sheldon's style, based on the prompt and the context retrieved from the vector database.

Use of RAG

1. Data Ingestion & Storage: The project ingests data from a CSV file and potentially other sources in the "data" directory.
2. Embedding & Indexing: The data is transformed into embeddings (numerical representations of text) using a SentenceTransformer model and stored in a vector database (powered by Pathway). This allows for efficient similarity search.
3. Retrieval: When a user asks a question, the question is embedded, and similar embeddings are retrieved from the vector database. These retrieved chunks of information form the context.
4. Prompt Augmentation & Generation: The retrieved context is added to the user's query to create a comprehensive prompt for the language model (Gemini Pro). The model then generates a response based on this enriched prompt.

By providing relevant context from the character's domain, RAG helps the model generate responses that align with Sheldon's personality and knowledge. Users might ask questions about specific episodes, theories, or events related to the show. RAG enables the chatbot to access and utilize this information to provide accurate and relevant answers. By retrieving relevant information, it reduces the risk of the model hallucinating or generating incorrect answers. RAG also enables the chatbot to be easily updated with new information by simply adding it to the knowledge base, without needing to retrain the entire language model.

Instructions for Setting Up and Running

1. Set up Google Colab:

   • Create a new Google Colab notebook.
   • Paste the provided code into the notebook.

2. Install Dependencies:

   • Run the following pip install commands in a code cell:

     !pip install llama_index
     !pip install transformers
     !pip install google-generativeai
     !pip install sentence-transformers
     !pip install pathway
     !pip install llama-index-retrievers-pathway
     !pip install llama-index-embeddings-huggingface
     

3. Mount Google Drive:

   • Run the provided code to mount your Google Drive.
   • Ensure the path to the Big Bang Theory script CSV file (csv_path) is correct.

4. Set Environment Variables:

   • Store your Google API key in Colab's userdata:

     from google.colab import userdata
     userdata.set('GOOGLE_API_KEY', 'YOUR_API_KEY')

5. Run the Code:

   • Execute all code cells in the notebook sequentially.
   • The chatbot will start running in a loop, allowing you to type in queries and receive responses from "Sheldon".

6. Environment Variables:

• GOOGLE_API_KEY: Your API key for accessing the Google Generative AI service. Store this securely in Colab's userdata.