RackGPT Chatbot Tasks
RackGPT - Goals, Fixes and Next Steps
Functionality Changes:
-
Fix session-specific chat memory -
Offer unique solutions for each user prompt given - consider previous URLs given -
Provide links to external support resources if chatbot is not providing good advice (Zendesk tickets, Slack community channel) -
Implement intent classification to inject custom chains based on user's input -
Implement vector-stored chat history (session-specific memory) -
Update RackGPT main repository README -
Write blogpost about RackGPT
Additional To-Do Tasks:
-
Push current version of chatbot to GitLab after fixing session-specific chat memory -
Explore different LLMs for better/faster inference and less API calls -
Chatbot Deployment -
Chatbot Evaluation -
Set up monitoring w/ callbacks
-
-
Implement simple UI using Streamlit
Future Implementations:
-
Text generation model implementation (RackGPT Preliminary Model) -
Integrate external knowledge bases for added expertise -
Wikipedia for facts on general topics -
Explore more external agents
-
-
File upload implementation - Requires plan for anonymizing files
-
Integrate DRP credentials for login - license auth - This will allow for saving multiple chat sessions per user
-
Deploy script (AWS Lambda?) that pulls docs, preprocesses, and vector stores (weekly?) to make sure documentation stays updated
Edited by Avneet Singh