Skip to content
This repository was archived by the owner on Feb 14, 2025. It is now read-only.

Milestones

List view

  • ## Goal Create a chatbot that can view images. ### Notes Previous research on creating this feature can be found in #199 ---

    No due date
    0/1 issues closed
  • ## Reasoning The question rephrasing is often a waste of token and time. It would be more helpful if the bot were ordered to drill in on the specifics of the user's question/concern to enhance the retrieval process. Currently, the chatbot "rephrases" the user's question every time, which often isn't rephrased at all, just repeated and used to query the retriever(FAISS Index). It would be far better if RAGbot "decided" when to rephrase the question, based on whether the user's query is largely incoherent, unspecific, overly verbose, conversation length (conversation recontextualization), etc.

    No due date
  • ## Goal Create a chatbot that retrieves the user's list of URLs, preprocesses them for only pertinent information, and context-stuff the information. ### Things to Consider - What is the optimal way to store the retrieved context, in Streamlit `session_state`, or perhaps a regular list? - How will the chatbot "know" when to refer to the retrieved context? - The user will be able to input multiple URLs, how will the chatbot "know" which retrieved context to refer to, if they are compartmentalized? #### Notes - Retrieved content should not have web-scraper noise, like HTML code, etc. - Guardrails for when the retrieved content should be used in the answer should be defined - Could use a LangGraph to direct the chatbot's "thinking" - Could append entire retrieved context to chat history

    No due date