AWS Machine Learning Blog

This post guides you through building a server for real-time PDF text extraction from Amazon S3 using the Model Context Protocol (MCP). It compares this interactive approach with Amazon Textract, detailing architecture, implementation steps, cost analysis, and security considerations.

Cara is an AI-native solution on AWS that automates back-office processes for insurance brokerages, saving agents ~10 hours per week. It uses Amazon EKS and Amazon Bedrock for scalable, secure, domain-specific AI workflows.

Stripe processes $1.4 trillion in annual payment volume across 50 countries. Using a ReAct agent framework on Amazon Bedrock, they reduced review handling time by 26% while maintaining human oversight. This article covers the technical architecture, infrastructure decisions, and lessons learned, including task decomposition, orchestration patterns, and cost optimization via prompt caching.

In this technical collaboration between AWS and the authors, we present a pragmatic solution: agentic overlays. Agentic overlays are thin wrapper layers that transform traditional REST-based services into agents capable of participating in A2A interactions. They also expose REST APIs as tools compatible with the Model Context Protocol (MCP). Together, they let enterprises add A2A capabilities to existing REST services without rewriting business logic, without duplicating code, and without running parallel infrastructures. This reduces agent sprawl in the infrastructure by reusing existing services as agents. We provide reference architectures and sample code that show how to build agentic overlays.

This post shows you how to configure training jobs on Amazon SageMaker AI to get the most out of Blackwell’s architecture on AWS. You learn how to select batch sizes and sequence lengths that take advantage of Blackwell’s expanded memory, choose the right precision format for your model size (1B to 64B parameters), and apply activation checkpointing strategically. By the end, you have a practical framework for tuning your training configuration and launching distributed training jobs on P6-B200 instances.

This post demonstrates how to implement video upscaling using SeedVR2 on SageMaker AI. It covers solution architecture, deployment steps, and performance comparisons highlighting quality improvements and processing efficiency. By the end, you'll have practical knowledge to implement this super resolution solution.

In this post, we show you how to build Chaplin (Customer Health and Planned Lifecycle Intelligence Nexus), an open source solution that uses AI agents exposed through the Model Context Protocol (MCP) to provide self-service health event analytics.

This post shows how to build a governed, serverless data mesh on AWS that provides the secure, scalable data foundation production agentic AI requires.

In this post, you will learn how to build a voice agent that handles appointment reminder conversations using Amazon Nova 2 Sonic and Amazon Bedrock AgentCore. The agent authenticates patients by voice, manages appointments (confirm, cancel, or reschedule), collects pre-visit health information, and escalates to human staff when needed. You handle routine calls at scale, which can help reduce no-show rates. This sample focuses on the agentic side of the problem: voice conversation and tool orchestration. A browser-based interface is included for testing. To connect the agent to actual phone lines for outbound dialing, you would integrate a telephony service such as Amazon Connect Customer.

This post explains how to build an end-to-end integration between Snowflake semantic views and Amazon QuickSight. Using movie review data, it demonstrates how to define a shared business logic layer, explore data with natural-language queries via Cortex Analyst, and generate consistent dashboards—reducing data reconciliation efforts and AI hallucinations.

Loka built a conversational AI agent using Amazon Nova 2 Sonic that addresses the latency and unnaturalness of traditional voice assistants, achieving high accuracy, low cost, and natural interactions through native speech-to-speech processing.

This post shows you how to build a conversational protein research assistant that combines three capabilities: Natural language query parsing to extract structured search parameters, vector similarity search over protein embeddings using a specialized language model and AI-generated scientific summaries of search results.

This post presents patterns for building production-ready multi-tenant AI systems using Amazon Bedrock AgentCore, demonstrated through healthcare AI agents serving multiple clinics and hospitals, covering tenant isolation, service tier differentiation, cost tracking, and observability.

Ampersend built a pay-per-intelligence routing layer on Amazon Bedrock AgentCore Payments, enabling AI agents to autonomously pay for model services using the x402 protocol. The integration handles wallet custody, spending governance, and two-hop settlement, reducing development time from months to under two weeks.

This post explores using multimodal embeddings, LLM captioning, and vector search on AWS to turn aerial imagery into a natural-language-searchable knowledge base. A five-stage pipeline built with Amazon Bedrock and Amazon OpenSearch Serverless evaluates different embedding models, fusion strategies, captioning approaches, and search methods. Experiments show Amazon Nova Multimodal Embeddings achieve the highest F1 scores on benchmark queries. The work evolved into Vexcel Intelligence, a searchable imagery product.

This post walks you through deploying ComfyUI workflows on Amazon SageMaker AI processing jobs to generate hundreds of high-quality images in a single batch. Learn to set up infrastructure using AWS CDK, configure GPU-accelerated processing, and automate image generation at scale. Adapt this solution to your own ComfyUI workflows.

Amazon Bedrock AgentCore now includes a fully managed web search capability that grounds AI agents in current web data. It uses Amazon's own web index, offers privacy by keeping queries within AWS, and integrates via MCP with minimal code. This solves the problem of stale knowledge in agents.

This article explains how to integrate Adobe Marketing Agent with Amazon Quick using Model Context Protocol (MCP). It details the setup process, authentication, and validation with sample queries for audience rankings, loyalty segments, journey usage, and conflict analysis, enabling marketers to get campaign insights via natural language conversations.

Amazon SageMaker AI now emits over 100 detailed inference metrics covering GPU health, token-level latency, KV cache pressure, traffic distribution across Availability Zones, and more. These metrics are displayed in a built-in SageMaker Insights dashboard in CloudWatch, which supports PromQL queries. This post explains how to enable detailed observability, navigate the dashboard, and connect metrics to external tools.

Amazon Bedrock AgentCore harness is now generally available, allowing developers to create and run a fully functional agent with just two API calls. It provides an isolated runtime environment, built-in memory, tool integration, skill libraries, and real-time tracing without writing orchestration code or building containers.

Today, we’re announcing inline payload support for Amazon SageMaker AI Async Inference. Customers can now send inference payloads directly in the request body of the InvokeEndpointAsync API, removing the need to upload input data to Amazon Simple Storage Service (Amazon S3) before each invocation.

Amazon Quick introduces autonomous agents that work continuously, a prioritized activity feed, and cross-data source insights from a single question, helping users reclaim hours daily.

At AWS Summit New York City, AWS announced a series of innovations including AWS Context (coming soon), AWS Glue Data Catalog Business Context and Semantic Search (preview), and Amazon S3 Annotations (generally available) to provide trusted context for AI agents. These services leverage knowledge graphs, identity-aware access, and open standards to enable organizations to build a shared, governed context layer that enhances the decision-making capabilities of AI agents.

Amazon Bedrock AgentCore introduces new capabilities for connecting agents to organizational, web, and paid knowledge, along with optimization features for continuous improvement and enhanced policy controls.

Today, we’re announcing a new API with Amazon Bedrock Guardrails. With this API, you can apply individual safeguards, also referred to as safety checks, at any point in your agentic AI applications without creating guardrail resources. In this post, we walk through how the InvokeGuardrailChecks API works and how to use it to build safe, multi-turn agentic AI applications.

Amazon SageMaker AI announces container image caching for inference, reducing end-to-end latency by up to 2x for generative AI models during scale-out events.

This post walks you through how to use P-EAGLE directly within Amazon SageMaker AI. It demonstrates how to select a compatible model from SageMaker JumpStart, configure parallel drafting, and deploy a highly optimized endpoint.

Today, we are announcing the availability of the Gemma 4 family on Amazon Bedrock. Built by Google DeepMind and released under the Apache 2.0 license, Gemma 4 is a family of open-weight models designed with a focus on intelligence-per-parameter across a broad range of deployment scenarios. The family includes three instruction-tuned variants: Gemma 4 31B, Gemma 4 26B-A4B, and Gemma 4 E2B. These cover dense and mixture-of-experts (MoE) architectures, where only a fraction of the model’s parameters activate per request. The variants offer built-in reasoning, native function calling, and multimodal input across text and image.

This post introduces detectors in the Strands Evals SDK that automatically identify failures in AI agent execution traces and perform root cause analysis, reducing diagnosis time from hours to minutes. You learn how to call detector functions, interpret structured output (categorized failures, confidence scores, causal chains, and fix recommendations), and integrate detection into your evaluation pipeline for automated diagnosis on every test run.

This post demonstrates building a competitive research agent using LangChain Deep Agents and Amazon Bedrock AgentCore. The agent delegates deep work to isolated subagents (browser and interpreter) to overcome context window limitations, enabling parallel research, data analysis, and cross-session memory.