With LLM in general they are great at summarizing data…
IF they know what the data represents.
I don’t know how you arrange your devices and sensors but for the whole lot of them the same thing applies.
If you were to list them, thier area, domain and alias in a big table and show that table to your grandmother with zero context.
What will your grandmother say?
(cause that’s basically where you’re at.)
In short it’s only as good as your ability to describe what it is.
You can use the alias field in any entity (where you select to expose to assist) and idescrive the thing.
Or you can write a description of what those things are in your prompt yes it’s a lot more detailed than that but if you ASSUME that’s your starting point you won’t give yourself poor expectations
I can get mine to tell me pretty much anything I want now but my user prompt is currently about 8 single space pages and burns out the conversation length in about 3 round trips on a chat (not ideal yes I’m actively optimizing and no i won’t share it it’s quite unique to my installation and has tons of PII data in it.and need a lot of work. What’s in it is not important just it CAN be done. If you work on it.) it’s also taken me nearly three full months of tuning
The art of prompt crafting (a number of industry individuals including myself are trying to avoid the term prompt engineer it’s not accurate) is tellig the llm exactly what you want in as few words as possible.
The problem you describe comes form the llm not having enough context. Your prompt and the resulting data set give the context. As llm models get better and transition into reasoning models you’ll see this get way better quickly - probably later next year as the reasoning models get to be en vogue.
Sorry for the general theory answer but you had a very theoretical question that a lot of people need to hear because they’re misunderstanding what the llm can do.
Remember right now unless you are in a lab building these things what you have access to is basically three steps above an overgrown autocorrect. It pattern matchds and tells stories like a banshee. Don’t expect it to be a wizard.