Preparing data for data agent

[Wickjone]

I'm currently exploring the capabilities of the data agent, one issue that i'm facing is that it doesn't seem to be able to understand the relationships between tables that well. Is it not able to directly read the relationships that i set in the semantic model within a lakehouse?

If i am unable to directly tell the data agent the relationships between tables, is it better to have one summary table instead? Or would giving the data agent instructions will suffice? If yes, what should the instructions look like?

Lastly, is there any way for the data agent to be able to query incomplete names? Examples is the name Amy Burns is in the semantic model, when i ask the data agent for "Amy Burns" the query is successful, however if i ask just "Amy", it will fail.

[v-ssriganesh]

Hello @Wickjone,

Hope everything’s going great with you. Just checking in has the issue been resolved or are you still running into problems? Sharing an update can really help others facing the same thing.

Thank you.

[Wickjone]

Hi, i would like to add on that adding the instructions to include wildcard characters for names is working as expected
what i did to improve the accuracy of the data is to denormalise the data to reduce to number of tables.

It made the data agent to process my queries faster and more accurate since it does not need to join many tables. 

[v-ssriganesh]

Hello @Wickjone,

We hope you're doing well. Could you please confirm whether your issue has been resolved or if you're still facing challenges? Your update will be valuable to the community and may assist others with similar concerns.

Thank you.

[DataGuru412]

Just wanted to share my two cents i tried  fabric data agent by providing relation in the Fabric data agent itself and it worked as expected. In some cases you will not be able to define each relation in semantic model but logically you can give instrction in Fabric data agent and it will work. I hope this might help you.

[v-ssriganesh · Accepted solution]

Hi @Wickjone,
Thank you for reaching out to the Microsoft fabric community forum.

The data agent can read relationships defined in your lakehouse's semantic model via its schema. If it's struggling, ensure clear column names and permissions. Adding specific instructions in the agent's configuration can clarify joins, like: Table 'Sales' joins to 'Customers' on CustomerID (LEFT JOIN, one-to-many).

Instructions are usually sufficient and more flexible than a single summary table. In the agent's settings, add table descriptions and Join logic. Example queries to guide the agent. A summary table is only needed for very complex scenarios.

The agent doesn't natively support partial matches (e.g: "Amy" for "Amy Burns"). Add instructions like: When asked for a partial name like "Amy", use SELECT * FROM Customers WHERE FullName LIKE '%Amy%'. Include example queries to train the agent.

Best regards,
Ganesh Singamshetty.