IS THIS setup overkill or not? [Solved]

How would you solve a case like this?

CASE:

I have a spreadsheet with customer questions in one column and a categorization of the customers' questions in a second column.

Once a month, I want my agent to create a report based on the information in the spreadsheet.

The questions are divided into 11 categories.

In the report, I would like to have a section/paragraph for each category along with the most frequently asked questions within each respective category.

Now, the spreadsheet can vary from a couple of 100s to 1000s of questions.

In each category section in the report, I would like to highlight the 10 mest FAQs.

Here are my thoughts:

❌ If I insert all questions from the spreadsheet into ONE AI-agent node (let's say 1000 questions), it will (in my mind) with 100% certainty fail, because it's simply processing to many questions and categories.

❌ If I make a filter and filter all questions by category and then based on the category activate a separate sub-flows (11 subflows because of the 11 categories), I will have to switch between 11 sub-flows when making changes or handling errors. This, in my mind doesn't sound like the solution.

✅ This is what I'm thinking - everything inside the same workflow. See the picture (It's not done).

Trigger --> Spreadsheet 1000 (or more) questions --> filter (by category) --> If statement (only passes 100 questions per category (to save token usage)) --> separate AI agents for finding FAQ's per category --> finally merging everything together into one JSON output.

AND, instead of running all agents simultaneously (and hit the TPM barrier) each agent will have a delay, so the agent before it is done analysing FAQ's within its category before the second initializes.

Do you guys have another, maybe more simple solution to this?

21 comments