[Bug]: Confused about the possibility of deep retrieval in a plugins without the external system being a vector database or supports semantic search.

[Raghavendra BN]

Sample

msgext-multiparam-js

Description

Plugin that demonstrates how to implement complex utterances and support deep retrieval.

I understand complex utterances with multiple parameters, but how is deep retrieval possible without ingesting external data into a semantic index or using Azure Cognitive Search or external system being some form of vector database? Are we not limited by search capabilities of APIs? In your example data source is simple json object collection, how does it support deep retrieval?

Steps to reproduce

Na

Expected results

NA

Actual results

NA

Additional Info

No response

[Garry Trinder]

Thank you for you question @ragbn

Are we not limited by search capabilities of APIs? In your example data source is simple json object collection, how does it support deep retrieval?

The sample you refer to is designed to show how a plugin can be created to support multiple parameters (compound utterances) through the configuration of the app manifest and show how to retrieve the values passed from Copilot in the query object when the plugin is executed. This is intentional. It is up to developers to implement the logic beyond this. In this case, "deep retrieval" does not mean RAG or Vector.

In the Northwind Inventory and Doc Search samples we show two different ways in which to implement the logic either by calling external APIs directly, or by using a RAG pattern and returning results based on that data back to Copilot for Microsoft 365 to use in it's response back to the end user.