siddharthp11 / rag-chatbot

RAG Pipline to automatically create, and talk to a vector database from data.

Built as a demo for the NYC NetCore Cloud office's content creators.

To create a vector DB:-

• create a directory in the root folder.
• within this directory create scrape_data.py
• within this module, write a function called scrape_data which formats and returns your data as a dict like:
  • {'title': [str], 'content': [str]}
• now, in the root directory open create_vector_database.
  • specify your directory name in DIRECTORY
  • enter your OpenAI API key in openai.api_key
• run create_vector_database. Since you are using openAI's API, you'll be charged for this!
• now, in your directory, you will see two folders:
  • 1. chunks.pkl are the chunks of text from your corpus.
  • 2. embeddings.csv is the embeddings database.

To interface with your data using the vector DB and an openai model:-

• In ui.py specify the {DIRECTORY}.
• Run streamlit run ui.py
• Enter your API Key.
• You should be able to use the chatbot and view the retrieval context that was provided to it.

Contribute:-

• OpenAI API flexibility-

  • User should be able to specify the number of retrieved articles with cost constraints.
  • Users should be able to see cost estimates.
  • Users should be able to choose an openai model.

• Reading the data files-

  • Users should be able to specify the data path on the UI.

• Creating the data files-

  • Users should be able to create the data files using some UI, where they can write the scrape data function.