Recap for April 30th, 2025 - Semantic Kernel Office Hours for US/EMEA

[sophialagerkranspandey]

Semantic Kernel Office Hours for US/EMEA - April 30th, 2025 recording: https://aka.ms/SKOfficeHoursRecordings

If you weren't able to attend office hours but still have questions, please post them here: https://github.com/microsoft/semantic-kernel/discussions

Meeting notes:

• Meeting Introduction:
• Sophia welcomed everyone to the Semantic Kernel office hours, highlighting the weekly and monthly meeting schedules and encouraging participants to use GitHub discussions for questions and issues.
• Recent Python Release:
• Sophia highlighted the recent release of version 1.29.0 for Python, which includes new features such as the ability to clone a kernel, support for process state management, and the addition of Brave search capabilities.
• Kernel Cloning: The new Python release 1.29.0 includes the ability to clone a kernel, bringing parity with the .NET version.
• Process State Management: The release also supports process state management, enhancing the control over process states within Python applications.
• Brave Search Capabilities: Brave search capabilities have been added to the Python version, allowing users to integrate Brave search via API.
• Recent .NET Release:
• Sophia mentioned the latest .NET release, version 1.48.0, which includes changes to Mebd, updates from Wesley and Shai, and various fixes.
• Mebd Changes: Sophia noted changes in Mebd, which were further discussed by Wesley and Shai during the meeting.
• Version 1.48.0: The latest .NET release, version 1.48.0, includes various fixes and updates, which participants were encouraged to review.
• Customer Case Studies and Blogs:
• Sophia discussed recent customer case studies and guest blogs, including an internal Microsoft adoption of Semantic Kernel for Microsoft Store Assistant and a demo on website accessibility analysis.
• Microsoft Store Assistant: Sophia highlighted a case study on the internal adoption of Semantic Kernel for the Microsoft Store Assistant, detailing their journey and usage.
• Accessibility Analysis: A guest blog demonstrated the use of Semantic Kernel for analyzing website accessibility, showcasing a practical demo.
• Microsoft Extensions Vectra Data Package:
• Weslie announced the release of the second preview of the Microsoft Extensions Vectra Data Package, highlighting improvements such as integrated embedding generation for search and support for dictionary objects for models.
• Preview Release: Weslie announced the second preview release of the Microsoft Extensions Vectra Data Package, which includes significant improvements.
• Embedding Generation: The package now supports integrated embedding generation for search, simplifying the process for users.
• Dictionary Objects: Support for dictionary objects for models has been added, enhancing flexibility in data handling.
• Integrated Embedding Generation:
• Shay explained the new feature of integrating embedding generation into the Microsoft Extensions Vectra Data Package, which simplifies the process for users by handling embedding generation automatically.
• Feature Overview: Shay provided an overview of the integrated embedding generation feature, which automates the embedding generation process within the Vectra Data Package.
• Implementation Details: Shay detailed the implementation, explaining how users can configure an embedding generator and integrate it with the vector store for seamless embedding generation.
• Search and Upsert: The new feature allows for automatic embedding generation during search and upsert operations, reducing manual intervention.
• Flexibility and Configuration: Shay highlighted the flexibility of the feature, allowing users to configure different embedding generators and manage embeddings without dealing with low-level details.
• Dynamic Mapping:
• Shay introduced the concept of dynamic mapping, which allows for the use of a dictionary instead of a strongly typed .NET type for scenarios where a fixed schema is not appropriate.
• Concept Introduction: Shay introduced dynamic mapping, which enables the use of a dictionary for scenarios without a fixed schema, providing greater flexibility.
• Implementation: Shay explained the implementation, where a dictionary of string to object can be used instead of a strongly typed .NET type, allowing for dynamic schema handling.
• Use Cases: Shay discussed use cases for dynamic mapping, such as tools or layers that need to work with any vector database schema defined by external configurations.
• Filtered Get Feature:
• Weslie discussed the new feature that allows for regular searches with filtering and custom ordering, which was previously missing from the Microsoft Extensions Vectra Data Package.
• Feature Introduction: Weslie introduced the new filtered get feature, which allows for regular searches with filtering and custom ordering in the Vectra Data Package.
• Implementation Details: Weslie explained the implementation, including the new getasync method that supports filtering and ordering parameters.
• Use Cases: Weslie highlighted use cases for the feature, such as ordering by timestamp to process records in a specific order.
• Collection Management:
• Weslie mentioned the addition of collection management methods, such as checking if a collection exists and deleting a collection, to the vector store object.
• Telemetry and Logging:
• Mahya raised a question about adding custom properties to logs in Semantic Kernel, and Tao suggested using parent activities and attributes to group logs by session or user ID.
• Custom Properties: Mahya inquired about adding custom properties to logs, such as session ID and user ID, for better monitoring and grouping of logs.
• Parent Activities: Tao suggested using parent activities and attributes to group logs by session or user ID, allowing for better organization and querying of logs.
• Follow-Up: Mahya expressed interest in following up offline for more detailed guidance on implementing custom logging properties.
• Multi-Agent Orchestration:
• Justin asked for recommendations on handling multi-agent orchestration in Semantic Kernel, and Chris suggested using a controller agent to decide whether to execute a plan or respond directly to user input.
• Orchestration Challenges: Justin described challenges with multi-agent orchestration, such as repetitive task execution by the planner agent.
• Controller Agent: Chris recommended using a controller agent to decide whether to execute a plan or respond directly to user input, improving orchestration efficiency.
• MCP Server Hosting:
• Raj and Jeevan discussed hosting MCP server as a containerized app and accessing it remotely, with Sergey providing guidance on using the MCP SDK and considering authentication and authorization.
• Hosting MCP Server: Raj and Jeevan discussed the challenges and considerations of hosting MCP server as a containerized app and accessing it remotely.
• SDK Guidance: Sergey provided guidance on using the MCP SDK for hosting and accessing the server, including setting up endpoints and handling requests.
• Authentication and Authorization: Sergey emphasized the importance of considering authentication and authorization, recommending the use of a gateway for security until the authentication story is fully developed.