Azure Open AI chat bot using csv source has issues in semantic search

Hello All,

I have a CSV Q&A source file which has issue number, description, solution and other fields.
I have done a vector query on issue description along with embedding. Semantic search is also incorporated.
Could someone help me on below issues? Any immediate help would be appreciated

  1. Issue number is a field in csv that has value such as “ISSUE00115”. User tries to use the chat bot by prompting explain issue 00115. Chat bot is unable to identify this in the csv even though it is present
  2. Similarly if user asks for certain category like retail and subcategory as diary, it sometimes does not consider the subcategory.


Have you considered putting those CSVs in a DB, and allowing the bot to construct a search query? :thinking:

1 Like

Actually your idea helped me ! I was able to achieve it using database as a source. But I am facing an issue. The response is returned in json format. How do I show it as a textual summary? Also how do I make the fields in select clause dynamic based on what the user is asking for?


the DB result? just give it to the LLM if it’s not too long and tell it what to do with it.

Just give the model your schema and tell it what it can do with it? :thinking:

Could you share any examples or blogs related to this? I am completely new and trying to figure this out :slight_smile:

None that I can vouch for, but it’s probably less complicated than you think. If you google Text-to-SQL you should find plenty of information :slight_smile:

1 Like

Found some videos online… Thanks!

1 Like