myamabot

An LLM based AMA bot, trained on data from my blog, website and Resume.

Setup

1. Install dependencies

pip install -r requirements.txt

2. Set up environment variables Put the following env variables (substituting for relevant values) into the .env file at project root.

portfolio_site = 'https://en.wikipedia.org/wiki/Sirius_Black'
blogs_site = 'https://harrypotter.fandom.com/wiki/Sirius_Black'
personal_blogs_site = ''
resume_url='<some path>.pdf'
openai_api_key = ''

3. Run the application

• Ingesting data: python app/indexing_data.py This ingests the data from the sources, storing embeddings in the InMemoryVectorStore first. Then the store itself is dumped to file system at the path specified in the .env (default: embeddings_dump.json).

• Checking generation: python app/retrieve_generate.py This loads the embeddings from the file system and uses them to answer questions (RAG workflow).

• Chatting with the bot: streamlit run app/app_streamlit.py Opens up a streamlit-UI chat interface on the localhosts

4. Deploying on Cloud

• I tried deploying on Google Cloud through the following steps:

  1. Build docker image (refer the Dockerfile in the repo)
  2. Tag the docker image such that it can be pushed to Google Container Registry. It would need to be of the format: HOST-NAME/PROJECT-ID/REPOSITORY/IMAGE
  3. Push the docker image to Google Container Registry using docker push <GCloud compatible image-name>
  4. Deploy the app as a container on Google Cloud Run.