ATOM

Customed chatbot based on prompted roles with Python, Streamlit, SQLite, and ChatGPT API.

• Chatbot 1: Generic by single prompt.
• Chatbot 2: Customer services with chain of thoughts.
• Chatbot 3: Allow for user to thumbs up and down
• TODO:
  • Disable like / not like when generate.
  • One question at a time.
  • Need to extract data with LLMs
  • Log feedbacks in database (Feedbacks)
  • Authentication for single user per feedbacks
  • How to allow manager to insert data (Azure Intelligence Docs? or Forms?)

Applications

• Use to train/assist multiple people with a small amount of documentation.

Ideas

• Create a form to submit "context", then store in database so that the bot has more information.
• Maybe create a customed form of attributes. Store in database.

Reproduction

conda create -n atom python=3.8
conda activate atom
pip install -r requirements.txt
cd src
streamlit run app.py

References

[1] Build a basic LLM chat app

[2] Building Systems with ChatGPT