I ve read through all the docs but I m

@xmalina ooc what's the data source? GPT Index is a general purpose tool over your text data. You can just feed arbitrary text documents into GPT Index, and then feed in an input prompt to get your answer

would it work for bank account transaction data?

I haven't tested it but would love to have you try it out and let me know!

out of curiosity what prompt are you thinking of feeding into it?

for example "which months did I spend above my budget for entertainment?"

as a type of example

but in a GPT context, each number will be a token, so in a big spreadsheet, it adds up fast

i think if you have tabular data it is better to put that in a structured DB like SQL and then show columns in the prompt and when a user asks a question like that, you generate the query

column names and probably data types for each column

Yeah this is a great point, related to some discussions here: https://discord.com/channels/1059199217496772688/1059587968567885946/1059864930821615698 and here https://discord.com/channels/1059199217496772688/1059350089581203536/1059921651627741195

There's a TODO for GPT Index to support structured formats by 1) supporting natural language to SQL, and 2) autoinferring schemas. I'm super excited about this direction. @xmalina in the meantime you could try a listindex over your spreadsheet data and seeing if that gives you adequate responses, but admittedly it will be a bit more expensive due to LLM calls

makes sense, and thanks @lucasneg for your thoughts. Will say in my use-case, it's not SQL data, think multiple dirty filetypes from different scrapes