[Crash Course] How To Build A Proper Knowledge Base

I get asked a ton about knowledge base and the outputs behind it - the answer is its basically a correlation search on information that is fed as context on the output. So if you have "bad outputs" or knowledge not being referenced, its probably because you didn't put enough data or "good enough" data for the embedding to understand.

So, here is a full knowledge base crash course and tutorial that shows you how knowledge base works, best practices, concepts, visualizations, and a walkthrough of a chatbot with no prompt but purely using knowledge to answer questions accurately.

📄 Documentation & Materials: https://docs.google.com/document/d/1AAWvAd0QYKDVYKIy8bUWgeJWazO7soQlKVNe75k_EDc/edit?usp=sharing

Video:

https://youtu.be/sFJDr-rPTLw

Notes:

Right now, voice doesn't pull knowledge - RAG is intensive and - traditionally - pretty slow. However in v2.1, we introduce knowledge into voice.
File uploads are halted and queued right now because we are going to restructure your information to be the most relevant on upload. This is why you dont get a "complete" status on files - FAQs, Text, and Scrape and fully functional at this moment.

5 comments