Category: Amazon SageMaker

In multi-turn reinforcement learning, your custom reward function decides what the model actually learns. This post shows how to design a composite multi-turn reward for Amazon Nova Forge, execute model-generated code safely inside it, and instrument each component to catch the pitfalls that quietly collapse a reward.

Learn how to combine OpenAI-compatible endpoints on Amazon SageMaker AI with Amazon Bedrock AgentCore runtime to build a multi-agent workflow where each specialized agent uses the model best suited to its job. This post also shows how to get token-level observability from SageMaker endpoints that Strands Agents does not instrument by default.

Learn how OneAdvanced, a UK enterprise software provider, built a UK-sovereign AI platform by self-hosting Llama 4 Maverick and Llama Guard 4 on Amazon SageMaker AI, with a RAG pipeline on pgvector and over 50 agents built with Strands Agents SDK on Amazon ECS.

Running large language model inference at scale forces a KV cache trade-off: oversized GPU instances or slow time-to-first-token. This post builds a tiered KV cache on Amazon SageMaker HyperPod that extends the cache into a shared, distributed NVMe pool with Curvine, so replicas reuse cache at near-local-disk speeds on cost-efficient instances.

The Amazon SageMaker AI Spaces add-on for Amazon EKS runs managed JupyterLab and Code Editor environments on the cluster your ML team already operates. This post shows how to install and configure the add-on, connect from the browser and from VS Code over SSH-over-SSM, and move your team to OpenID Connect sign-in with Amazon Cognito.

The Amazon SageMaker Python SDK v3 now exposes generative AI inference recommendations in Amazon SageMaker AI directly in your notebook. Benchmark an endpoint, generate data-driven deployment recommendations, and deploy the recommended configuration without leaving your notebook workflow.

Learn how to build an inference meta-monitoring system for Amazon SageMaker AI endpoints using Amazon Quick. This governance layer sits above production ML inference pipelines to continuously track prediction and data quality, detect drift, integrate delayed ground truth, and surface automated performance dashboards.

In this post, we cover why Deepgram built on IAM temporary delegation, how the integration works end-to-end, and what it unlocks for customers running Deepgram speech models on SageMaker AI. With this integration, Deepgram has reduced the time for initial investigation on a SageMaker AI support ticket from days to minutes.

Learn the architecture and design decisions behind an explainable next-best-product recommendation system for banking, built with Amazon SageMaker AI and PyTorch. A multi-tower neural network with learned attention delivers accurate, per-customer recommendations while providing the explainability that banking regulators require.

In this post, we explore an idea for generating thinking tokens for datasets that lack reasoning traces in SFT customization. We first examine the reasoning suppression problem, then introduce Self-Distilled Reasoning (SDR), validate it across three benchmarks, and provide practical recommendations.