Lindy Community Icon

Join the Community

Optimizing AI Agent Structure for Efficient Research and Reporting

·

·

I'm trying to set up research functionality in a lindy and i was struggling using a single AI agent with access to all the platforms and"skills" i needed so i ended up deciding on a structure where a research coordinator agent works with a bunch of researcher agents to prep a report and feed it back to me Executive Assistant Lindy agent to draft a summary. Is this an insane structure? haha. My main issue here is getting the coordinator to call the different researchers in on research requests and then there's this constant issue with completions getting registered too soon and processes racing ahead while a RAG search is still running or several tiktok scrapes are still processing. I'm struggling to tackle these within my agent prompts and it's leading to information either getting missed or multiple emails being sent. This will be critical because i'd like to eventually train my EA agent to be able to then summarize these research reports into something executable by a graphic design agent to then hopefully send back a visual like a deck or info graphic that i can the give feedback on (hopefully not needed haha) to complete a process. the end result should be comprehensive concise reports in emails or visualized representations of reports i can share with my team. any thoughts on my structure here or how to account for any of these race ahead problems?

11 comments

Marvin A.
·
·
Is this an insane structure? haha.
yes. Lol 😄 jk I’d try to separate the agent from crucial or long-taking executions like RAG or scraping. Generally speaking I like the way you approach this but it needs some more logic to make it more reliable
Marvin A.
·
·
lmk if this is helpful!
Marvin A.
·
·
also —what utilities did you give your EA lindy??
Crispin D.
·
·
The EA has Modify Lindy Memory and Add Context utilities. Do those make sense? I'm learning as i go and burning through so many credits it's hilarious but i'm enjoying it and want to become proficient at building agents in Lindy!
Crispin D.
·
·
Ok so would you go about this with a research coordinator agent like in the screen grab and then attaching each of the tools so that they are engaged whenever the research coordinator agent wants to engage them and then they return once processes are complete? Would you group tools together in sequence in some instances like "Extract Webpage Text" after "Search The Internet"? like is there logic here where the extract webpage text tool needs to have access to the internet first to do it's scraping? after that would you attach all the tiktok and instagram tools etc similarly?
Crispin D.
·
·
also i am sort of making a giant super agent at the moment which could potentially go up in flames. creating a very purpose built Executive Assistant that can work across 3 company google calendars etc
🤯1
Marvin A.
·
·
The EA has Modify Lindy Memory and Add Context utilities. Do those make sense?
That’s dangerous terrain 😄 not sure how well the LLM will manage it’s own memory and context I’d avoid that for sure.
✅1
Marvin A.
·
·
like is there logic here where the extract webpage text tool needs to have access to the internet first to do it’s scraping?
Yes!
Marvin A.
·
·
First search → so it has access to url’s, then extract → relies on the url
Marvin A.
·
·
also i am sort of making a giant super agent at the moment which could potentially go up in flames. creating a very purpose built Executive Assistant that can work across 3 company google calendars etc
Thats crazy —and you just started using lindy? super cool
Crispin D.
·
·
yes i'm basically not sleeping and just testing new ideas for my team hahahaha so so cool and fun but also totally mind bending but am determined to master the logic and flows becuase i can see endless use cases
🔥1

Optimizing AI Agent Structure for Efficient Research and Reporting | Lindy Community