I have my custom Agent working well with a dashboard within couple of sheets contains tables and line charts.
however, when i gave a dashboard with two table sheet, and the table’s dataset contains 1M rows of 148 cols in SPEIC. but the Agent said it has technical issue to use that dashboard. i have also try to ask the agent to filter on the data first but yet to work.
so how does the agent filtering data from the dashboard? does it intelligently enough to implement filters ? or it would understood run queries to the dataset behind the dashboard? thanks!
Hi @longzhiw1 and welcome to Quite Suite Community!
To first answer your question on filtering data with an agent, I don’t believe that is possible at the dashboard level. Agents can answer questions using the data sources they’re connected to, but they can’t apply dashboard filters or interact with dashboard’s UI. And if the dashboard is the only component attached to the agent, then it also isn’t able to run queries on the underlying dataset.
For the technical issue you’re running into, my guess would be that the dataset is too large or too wide for the agent to interpret when it’s only being accessed through a dashboard. Agents work best when the data is structured inside a space or a topic, where the semantic layer is explicitly defined. In that context the agent should run normally, but when the agent relies solely on a dashboard, especially one with multiple table sheets and a very wide dataset, it may not have enough semantic information to operate, which often results in technical errors.
@JacobR Hi, i did a bit search on such, but looks like not too many docs described similar request. a very close on is saying topics might be an option. do you have any further doc could help? ty
I’m glad my previous post answered your initial question!
It entirely depends on your use case, but Topics or even Scenarios could be what you’re looking for. For topics, it is essentially like a Q&A chat box where you can ask questions and the AI model will interact with your dataset (perform operations such as filtering rows, aggregation, etc). Because you directly connect your dataset to it, you should not run into the same problems as you did before when you tried to ask questions to your agent when your agent was only connected to the dashboard itself. Now for topics it can get a little tricky as it does require a good amount of set up to make sure the fields are being properly called on when a question is asked. And it does require a bit of ‘training’ to set up in regard to getting the right answers to your questions, but nonetheless it can be a very valuable asset once it is properly set it. For more details I would refer to this resource as a start point (Topic workspace - Amazon Quick).
For scenarios, it is similar to topics in the sense that you can ask it questions but different in terms of its uses. Scenarios are more for “what if” type of questions if that makes sense (i.e. if I discontinue Product A, how will that affect profit margins). Unlike topics, it does not require as much of an effort to set up. Additionally, you can start a topic by selecting an individual visual from a published dashboard (i.e. a pivot table/table) or you can upload a dataset from your computer. Refer to this resource as a starting place as well (Working with scenarios in Amazon Quick Sight - Amazon Quick).
Both have their own use cases but are extremely valuable in terms of the insights you can get from them. Hope this helps!