How I started GrantReporting.ai

 Initially I've talked with my wife, she described me the process. 

I tried a few prompts using different models, actually all that was available at that time:
Gemini, Claude, ChatGPT, Perplexity, Grok, Manus, DeepSeek 

As an input, gave them all the documents, asked to create me a report.
Of cause wasn't perfect at all, but good signs that it's possible to do it, I received. 

So next I tried to be more specific. 
By the end, choose Gemini to work with. 

Initially I was using a big prompt, in 1 step to transform the documents into a report. 
Very Very buggy and consistency was lacking. 

So split it into 4 steps, and for each one created a Gem. 
Worked pretty well. 

BUT

it was the theoretical , my vision, approach. 

In reality I had a problem: My potential users, was using Google Sheets, Docs, Gmail etc. 
The Reports are made in Google Sheets predominantly, shared via Email.

So at least for now I need to keep Google sheets. 

Another problem is that Sheets needs to be flexible, I mean the Reporting Template can have 1 row, but the Report itself, for one client can have 3 rows for another 50 rows, depending on how many transactions they made. 

After playing around, I decided that the Gemini based on the received Reporting template will create a Google Aps script to re-create from scratch the Template + 
via the script, will update it based on the other documents.