This session presents Agentic Datasets, a cloud-native pattern where datasets act as intelligent, self-managing data products on AKS. Each dataset runs as a DatasetAgent Custom Resource encapsulating metadata, reasoning logic, and executable workflows defined through a lightweight SDK. Workflows are distributed as signed OCI bundles and executed securely within the cluster, enabling automated validation, embedding generation, and lineage tracking without external orchestration.
Using controllers and GitOps pipelines, Agentic Datasets integrate seamlessly with service mesh, observability, and governance frameworks. The poster illustrates how datasets can reason about their own content, apply LLM-assisted automation for summarization or decision tasks, and participate as active, autonomous workloads in AI-enabled Kubernetes environments.
The design integrates with Azure OpenAI Service and Azure AI Foundry for LLM-assisted reasoning, enabling DatasetAgents to perform contextual analysis, summarization, and autonomous decision tasks directly from within Azure-hosted workflows.