I recently have dove into the AI sector of the app, its awesome to say the least!
One road block I am encountering is I am trying to feed specific data to the instruct portion in order to return very specific data from documents. More using it as a comprehensive document search engine. I have combined multiple PDF files into a word doc and then pipe that file into the AI generate text instruct section. However some of my files are too large and are exceeding the character limits of the AI
How are you guys overcoming this obstacle without multiple columns of different data. I could categorize the data and have smaller chunks of data, but this involves more user input which is always bad and leads to errors.
Sorry I should have been more specific, I am using the document to Text, then I feed the output from that to the input of the other AI generate text prompt
I have figured this out, What I Did was create an openAI assistant (this is not free), I trained the assistant on my larger data documents that are combined into one document. Then I will use a webhook from glide to send to the assistant via make.com, then I will directly edit the glide row I want to see the response in.
waiting on glide to finish the make.com integration before I can finish up this automation.
Yes essentially (untested because the glide integration on make.com is currently down)
Setup user input prompt in app through text input field. Upon submit send webhook to make that trigger ls your assistant to run with the prompt. Assistant already preloaded with data needed and instructions. Then send the output of assistant to an edit row function to display in the app.
Chatgpt4 turbo seems to be working well albeit not super cheap on my 60,000 token documents. But very effective at pulling information and telling users what to do.
