Semantic Kernel plugin

[ross-p-smith]

Motivation

To demonstrate how and when to use semantic kernel as an AI agent.

How would you feel if this feature request was implemented?

Requirements

• Add a new orchestration strategy that leverages semantic kernel
• Document how and when to use each orchestration strategies
• Follows a sample like semantic-kernel/python/samples/kernel-syntax-examples/azure_chat_gpt_with_data_api_vector_search.py at main · microsoft/semantic-kernel (github.com)

Tasks

To be filled in by the engineer picking up the issue

• [x] #757
• [x] #758
• [x] #766
• [x] #759
• [x] #867

[reechamishra]

@gmndrg Hi Gia ! Can you please elaborate the requirements and more details to this task ?

[gmndrg]

@mrcarter8 Can you please help @reechamishra who will be working on this feature, with the requirements for the Semantic Kernel plugin? Thank you so much!

[cpich3g]

@reechamishra The idea is to build an AzureAISearch plugin using SK and then leverage it similar to the vector search plugin build for CosmosDB. There are few samples available that are being used as base for this (https://github.com/Azure/azure-search-vector-samples/blob/main/demo-python/readme.md). Since SK has released a new version for their SDK, working on building this using that.

[mirojs]

@ross-p-smith While developing this nice feature, it would be even nicer to keep an eye on a simple way of including TaskWeaver as well. Thanks.

[reechamishra]

@mrcarter8 Can you help us with requirements on this issue ? We would like to pick this one to work on next sprint if we get more details for engineers to work on .

[cecheta]

The first pull request to add Semantic Kernel to the repository has been merged: #758

For the QuestionAnswerTool (RAG), LangChain is still currently being used at the moment. Using Semantic Kernel memory for RAG does not seem to be suitable for this particular use-case because:

• The AzureCognitiveSearchMemoryStore class does not (at present) allow you to configure the names of the fields in the Azure AI Search index
• Due to @adamdougal's ongoing work with adding computer vision embeddings, Semantic Kernel does not support multiple embeddings (neither does LangChain)

For now, we will look to use the Azure AI Search SDK to perform the retrieval of documents, however we can use Semantic Kernel to call the LLM after the documents have been retrieved.

[cecheta]

Moving to "Planned" until work to remove LangChain from tools has been implemented as part of #323.

[gaurarpit]

I think this is done. Can we move it to "Done" @superhindupur ?

[gaurarpit]

All associated tasks have been closed.

[superhindupur]

I think this is done. Can we move it to "Done" @superhindupur ?

Yes!