Does anyone have an idea of how to use CSV files in the KB? Maybe a API
Hey, so in my project, im trying to use CSV files as my KB so that the bot can provide insigth on data. But per now, I know it’s not supported in VF, and I was wondering how I maybe can work around this. Does anyone have some tips or suggestions? Any help appreciated💪
24 Replies
To implement the Knowledge Base in Voiceflow, you can start by navigating to the Knowledge Base tab within the Content section of your dashboard. Here, you can add new data sources by clicking the 'Add Data Source' button. The types of data sources you can add include URL(s), Plain text, Sitemap, PDF, and Docx.
Once you've selected the type of data source that aligns with your requirements, you can set a refresh rate for the data source. This ensures that the information remains up-to-date. You can choose from options such as 'Never', 'Daily', 'Weekly', or 'Monthly' based on how frequently you want the information to be updated.
You can also perform bulk updates, which is particularly useful for managing multiple data sources simultaneously. This feature can save you time and effort.
To further refine the performance of your Knowledge Base, you can adjust settings such as Model, Temperature, Max Tokens, Instructions, System, and Chunk Limit through the KB Settings modal. This allows you to tailor the generation of responses to your specific needs.
For more detailed instructions on how to upload documents and manage your Knowledge Base, you can refer to the following resources:
- How the Knowledge Base Works
- Step 2 - Add Data
Please note that currently, Voiceflow does not support CSV files directly as a data source for the Knowledge Base.
What kind of data are you looking to store?
Mostly research data and number prognosis
The issue is that the KB is built to minimize token usage and to grab relevant data to produce an answer. This is the case for all KBs, not just VFs. It does not work the same as ChatGPT and uploading a doc for it to analyze.
This does not mean it is not doable. How much will it cost you in token usage? Well, that is another story.
You might look to integrate an OpenAI Assistant into VF. Again, the cost will be significant.
Rigth, but is There another tool that you know of that migth be avle to use CSV effectively and cost effectively?
Never done that before. Is it expensive to maintain?
foreign-sapphire•15mo ago
just an idea:
what if you convert your csv to json?
you could then use voiceflow and a set amount of scripts to interact/filter with and search through your json (that would only be necessary if you have a particularly large csv otherwise you could simply send the entire json of the csv to the llm everytime you have a question)
this has piqued my interest, im going to look at it now
Interesing idea, not quite sure how i would do that. Can you send me some tips if you figure it out?
foreign-sapphire•15mo ago
quick question, this csv is information saved in the backend correct? it is not automatically updated?
Yes, its from a client, but i think so. Does it matter?
foreign-sapphire•15mo ago
well if its always the same it will be very easy to work with, if it ever needs to be updated or changed that adds a bit of complexity.
Makes sense, this could be a cool project
foreign-sapphire•15mo ago
could you dm me an example of what your csv looks like?
raw and delineated, not a picture or anything. should look somethting like this:
Username; Identifier;First name;Last name
booker12;9012;Rachel;Booker
grey07;2070;Laura;Grey
johnson81;4081;Craig;Johnson
jenkins46;9346;Mary;Jenkins
smith79;5079;Jamie;Smith
Sure, il try with an example.
foreign-sapphire•15mo ago
https://creator.voiceflow.com/prototype/6625427b65bcc6956845e717
here's waht i came up with real quick, shows a couple different options.
basically we are just storing your database as a variable and referencing it that way, theres a couple different ways to do it, but mainly you jsut need to utilize code blocks to actually interact with it. technically you do not need to convert it to json but in my opinion it's easier that way.
Cool, il definitely try it out
Thanks! I’m not that good at coding, but thank you for explaining it to me
foreign-sapphire•15mo ago
let me know what you think.
i just saw the transcript, you inputted raw csv which was delineated by ';' i only have it setup to convert csv delineated by ','
i can adjust it, one sec
for now just use a csv delineated by commas
should work now with both semicolons and commas
Great! Thanks, I’ll try it out in a sec
You can use vectorshift via api for csv, https://vectorshift.ai/
https://www.youtube.com/watch?v=fhCW_vZW3Qc
Sounds cool, how is the token consumption?
This will still cause issues. The KB will still not pull all the correct info. This is because of the chunking and trying to minimize token usage.
foreign-sapphire•15mo ago
im under the opinion that when possible a programmatic approach is not only the most efficient but also the most accurate. any AI usage should probably be reserved until the very end in most cases.
So is there a way for my clients to be able to put in data in the KB without touching VF?
@Radiant raindrop DM me
I need more info
Very cool solution actually! Thanks for being so helpful🙂