Weekly Machine Learning Roundup: Agents, HPC, and In-Database Search

This week’s machine learning articles focus on efficient AI-driven workflows for industrial and research teams. There are updates on in-database semantic search, agent frameworks, industrial deployments, and research in the life sciences. Tutorials and case studies provide actionable examples and show practical adoption of advanced ML tools.

Building AI Workflows with Microsoft Agent Framework and .NET AI Stack

Building on recent themes of local embeddings and agent-based architectures, Pamela Fox’s livestream series demonstrates using the Microsoft Agent Framework in Python, with coverage of RAG agent skills, modular and reproducible AI deployments, monitoring using OpenTelemetry, and orchestration via Magentic. Evaluation with Azure AI SDK rounds out the workflow. .NET developers can join the AI Community Standup, which now features hands-on sessions using Semantic Kernel, AI Extensions, and orchestration tools—helping the .NET community move beyond chatbot projects to deeper AI integration.

Industrial ML and Scientific Workflows Powered by Azure HPC and Microsoft Discovery

New case studies highlight large-scale machine learning on Azure’s HPC resources, such as Neural Concept’s industrial engineering work with Azure GPUs and storage for AI training in automotive aerodynamics. Benchmarks show efficient model development that parallels what was seen last week in deep learning rollouts. In drug discovery, Insilico Medicine’s Nach01 model deployed via Microsoft Discovery demonstrates secure, repeatable analytics in the life sciences, drawing on Azure ML’s compliance and deployment features.

Expanding Vector Search in Databases: DiskANN in Azure SQL and Fabric SQL

DiskANN now enables large-scale, fast vector search directly inside Azure SQL and Fabric SQL, building on last week’s announcement of local-embedding in the Fabric Eventhouse. This lets teams implement semantic search, classification, and content analysis at the database level for less latency and stronger privacy, without relying on outside APIs.