Single Index vs Multi-Index for Mixed Data Sources, Best Practice?

[NormalNickYT]

So, I’m currently working on a RAG chatbot using this repository. I’ve already implemented a SharePoint pipeline that captures all changed content daily, updates Blob Storage, and applies the necessary index changes to keep our chatbot up-to-date. This setup is working really well and is already delivering great results within our company. Thanks for this amazing repo❤️

Now, we want to scale up with additional data sources. So far, we’ve been working with documents, but we also want to include a new source: CMDB/EAM content. You can think of this as SQL-like data containing tickets created in our company, with information such as ticket ID, title, solution, status, and other related metadata.

I’m looking for advice on how to structure this. My idea is to create one index that includes fields for both the SharePoint and the CMDB data source. Metadata fields specific to the CMDB would be populated when the document comes from the CMDB, while SharePoint fields would remain null, and vice versa.

For the content field, SharePoint entries would use the document chunks as usual in this repo, while the CMDB entries would use a concatenated string of all relevant metadata for example:

"content": "JobId: …, Title: …, Description: …"

This allows the field to be used for embeddings.

I plan to use Agentic Search to handle more complex queries, especially for the CMDB content, as users will likely ask multiple questions in a single prompt.

My question is: does this approach seem appropriate? Or is combining different types of data sources in a single index not recommended? Would multi-indexing with a custom orchestration or even multiple chatbots be a better approach? What is considered best practice for a scenario like this?

[cforce]

Either vector search is supported via several indexes (in same azure search service and with same schema) or not.
If not - several index would have a major disadvantage
If yes why not in general maintain each category in own index. Question: Which features would this enable which is not possible by using the same index.

e.g

• Effort to drop add new document categories
• Effort to backup and restore and prune single categories (indexes)
• Access rights management and secure boundaries for one index
  ....

[pamelafox]

Azure AI Search agentic retrieval feature now supports sending queries to multiple indexes:
https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/agentic-retrieval-updates-in-azure-ai-search/4450621

We will be adding support to that in this repo, along with several other datasource-related improvements. We aim to have that support added before the Ignite conference in November.

I think separate indexes are best if the data is semantically different- like there are clear categories of questions that would never need one index.

[cforce]

@pamelafox
It would be great to introduce a feature that stores either a category in shared or in dedicated index.