Microsoft Semantic Kernel, “the core of Copilot”, to combine your data or documents with the ChatGPT, Bing and 365 Copilot engines and build powerful applications.

Microsoft Semantic Kernel, “the core of Copilot”, to combine your data or documents with the ChatGPT, Bing and 365 Copilot engines and build powerful applications.

The Semantic Kernel, also known as the Semantic Kernel, was specifically created to allow software designers to easily integrate artificial intelligence services into their already existing applications. For this, the semantic kernel provides a series of connectors that simplify the addition of models and memories, thus providing a kind of simulated “brain” to the application.

It is available for C# and Python languages. It is open-source. The engine is still young but already very interesting to use. It does not yet have the functionalities of a Langchain (for Python) but it will evolve very quickly.

Our shellbots tool, in addition to Langchain, uses this library to use personal data, documents or computer code with AIs on the market to generate code, services, applications related to the customer’s context. Example: A client wants to build a natural language search engine on data from an ERP like Dynamics 365 Business Central. For this, we will mirror the ERP data in a vector database and then process it with Microsoft Kernel Semantic. We could ask the engine: “Give me the TOP 5 sellers this week with their turnover” in connection with the data contained in the application. Or also obtain information in the ERP and then automatically ask a question to OpenAI or Bing on this element for example.

Additionally, adding functionality to your applications is made easy with Semantic Core AI plugins, which give you the ability to interact with the real world. These plugins consist of triggers and native functions that can respond to certain actions. In a way, these plugins act as the “body” of your AI application. We can thus integrate in an application, the possibility of asking questions on a complete pdf file for example or on a vectorial database. With OpenAI and ChatGPT for example, it is not possible to integrate personal data but thanks to the semantic kernel, we can combine the 2 to create powerful applications.

Thanks to the extensibility offered by the semantic core with its connectors and plugins, you can use it to orchestrate plugins designed for ChatGPT, Bing and Microsoft 365 Copilot on models from OpenAI, Azure or even Hugging Face.

As a developer, you have the option of using these components either individually or in combination. For example, if you only need an abstraction of OpenAI and Azure OpenAI services, you can use the SDK to simply run pre-configured prompts in your plugins. However, the real power of the semantic core lies in the combination of these components.

But why do you need an AI orchestration SDK? You could choose to use the APIs of the most popular AI services and directly feed your existing applications and services with the results. This would, however, require you to learn the APIs of each service and integrate them into your application. In addition, using APIs directly would not allow you to easily take advantage of the latest advances in AI research, which require solutions based on these services.

By using multiple AI models, plugins, and memory all together within the Semantic Core, you can create complex pipelines that allow AI to automate complex tasks for users.

For example, let’s say you use the Semantic Core to create a pipeline that helps a user send an email to their marketing team. Using memory, you could retrieve project information and then use the scheduler to auto-generate the remaining steps using available plugins (e.g., use Microsoft Graph data to understand user request, generate a reply with GPT-4, and send the email). Finally, you can return a success message to your user in your application using a custom plugin.

To sum up, the Semantic Core is a valuable tool that makes it easy to integrate AI into your existing applications. With its connectors and plugins, it allows you to orchestrate AI plugins from various vendors. You can choose to use these components individually or together, depending on your needs. The Semantic Core is here to simplify the process of creating AI applications and to help enterprise application developers integrate AI into their existing applications.

Source: Microsoft. See documentation


Leave a Reply