Knowledgebase
Hey!
We have multiple solutions in Voiceflow, but yet we think that one of our main issues is how to operate Knowledgebase. Sometimes it feels like the AI Agent step does not fetch Knowledgebase information correctly.
Does anyone have any good links / videos how to utilize knowledgebase correctly? For example, when asking AI Agent for Roof details, the AI would fetch the information directly from a PDF -file in knowledgebase
We stumble on issues quite often, when the data and key information is inside the knowledgebase, but for some reason the AI does not fetch the information.
Is there some crucial trigger texts that has to be imported to the AI Agent inside prompt? Or how it should be structured. Any videos / links / help texts are super appreciated! :)
1 Reply
hello @Svimu I totally get what you are asking about..KB is a whole topic in itself...just recording video tutorials about how to optimally operate KB in Voiceflow.
I myself have needed to come up with a way how to structure KB optimially in order to achieve at least 90%+ RAG accuracy, which took me quite a while. ...I can give you couple of advices for starters:
1) Data structure: make sure your KB is well structured in atomic chunks, ideally in tabular format (Q&A format / Product tables, etc.) enriched with metadata and keywords, so that you can query and filter the KB in best way possible
- this part is the hardest and often needs to custom code the scraping/data pump via api (we have created our own software for us to do this)
2) Construct metadatafield filters dynamically via LLM based on the user query in order for you to narrow down the KB data set and extract only the most relevant records
3) Query optimization - You need to optimize the query into a RAG / vector search version enriched with most pertinent keywords based on the records you are looking for
Shoot me DM if you wanna chat more in depth (y)