How to efficiently reference file content in cloud, reducing local copies?

Hi community,

I've been building an ICM for my job as a Solution Consultant. One workflow I have is about status updates on customers that require referencing the latest in MS Teams chats/channels and Sharepoint files in a few folders (word, pptx, excel, mp4), among many other tools. In this post I will focus on those two.

Workflow descriptions:

Workflow that runs the first time I ask about a customer, which searches everything and build the first 'log' for that customer. This one I'm happy with.
Workflow that runs on any subsequent request that searches the latest 2 weeks of content, or the latest since the last 'log' on the customer. This one I need help with.

ICM structure:

CLAUDE.md: Routes the request to the customer folder context file
/customer/CONTEXT.md: Describes the two workflows and selects which ones to run and steps on how to create the 'log' files in new subfolders. Each workflow references the reference context file to know how to connect to tools like cloud storage to pull content.
/references/CONTEX.md: Has all the descriptions on how to connect to all cloud tools and where to look for info, like hyperlinks to customer notes folders

Question:

The problem I'm running into is that sometimes a key fact is missing from the response because of the way the information got parsed by the LLM. I'm not expecting it to read everything, but I do expect it to prioritize certain things when I'm updating how I want the status to look like.Are there some tips you can share with me?

I was looking into knowledge graph, but the examples I saw were using files locally. Maybe I need to markdown-ify my content? I want to hear from the community if you have ran into something similar.

Thank you!

2 comments

How to efficiently reference file content in cloud, reducing local copies?