I've been working on moving our AI logic to Retool Agents so I thought I'd share the flowchart (MS Paint lol) of what we're trying to accomplish and see if anybody had any thoughts, opinions..... improvements?
I tried to organize it the best I could where the answer generation is in the bottom part and the response formatting and tools for replying to external services are on the right.... and Sanitizers/Accuracy guards are in the middle (Input Cleaner Agent, Response Cleaner Agent, Response Fact Checker)
On a random side note, does anybody happen to know how deeply nested recursive Agent calls can get? My 'Fallback' and 'Fact Checker' could both cause some really deep recursive calls under the "perfect" (un-perfect?) conditions