Simplest , Build a chat interface to build queries for tcl.  Example find all orders  for loyalty members  customers  who have shopped  in last 30 days with gift cards purchased  in the last 90 days . Currently this requires creating  creating  multiple save lists to get the data . 
In this example we have 3  datasets . Transaction   history  .  Gift card , loyalty customers .  I  eventually gave up exported all  dataset in excel  and did lookups . Would have been faster if I could have generated all the tv queries to save list and use them up filter data  via gpt style ai chat bot 

Hi Mike, 

Please check out my article at Medium ( might need a free login ). I will be publishing several open source training blogs and publishing the code on github. 

https://medium.com/@kkonchadi/chat-with-your-u2-unidata-files-locally-by-using-open-source-llm-creating-a-rag-using-u2-unidata-c96d8527979c

Thanks in advance for starting this thread!

Regards,

Kurt

That's amazing, and I plan on working through your examples in the demo account, thanks for putting together such a detailed walk through!  Interested with experiences doing it at scale - if you have tables with millions of records, does it still hold up?

I was thinking of a customer service chat app, where a customer could ask questions about their account, orders, billing etc. However, it seems like a difficult problem to restrict an LLM to a subset of topics - the idea of a customer doing an LLM jail break and being able to trick a chatbot into revealing other customers account information is a security nightmare...

Hi Ian, 

Thanks for going through the article and appreciate your feedback! 

The current article was intended to demonstrate, how to interact with Unidata tables using a Large language model and provide textual responses to your questions.

Thank you Kurt.   This is a good beginning to see where AI will take a database of a related subject.