Hi,
I want to build a workflow which will classify new transactions from my bank accounts into expense categories, based on prior examples i’ve done manually.
The new transactions are in a google sheet, and the data includes information such as the merchant, date, which card was used, amount etc. These fields could be in multiple columns, or i could concatenate them. Of these, the merchant and card used are probably the most predictive.
I have around 3000 prior transactions classified in another sheet. These classified transactions have the same info as the new transactions (merchant, date, card, amount), but they also have one of ~40 different expense categories allocated to each row.
So i want the AI to use these prior transactions as examples to classify the new transactions. This will be a fuzzy match problem, not a pure lookup. I only want the classification if the AI is quite confident. If its not confident, then i want to return “Unknown” and i’ll manually classify them.
I’m wondering if this is possible? I’ve tried, but the challenge i come across is what workflow step do i use to upload the 3000 prior transactions into the context window so that AI can classify new transactions? I only seem to be able to generate prompts where it needs me to list all the examples into the prompt window which is not practical given there are 3000 examples.
Any help or suggestion is greatly appreciated! Thanks!