From AWS Machine Learning Blog · 40 stories
Amazon SageMaker HyperPod Enhances LLM Inference with New Features
Amazon SageMaker HyperPod has introduced new features to enhance large language model inference, including Disaggregated Prefill and Decode, support for Hugging Face, NVMe integration, and model quantization with Unsloth. These updates aim to improve performance, observability, and resource efficiency for enterprise users.
BYOKG and GraphRAG initiatives enhance pharmaceutical drug discovery
Pharmaceutical researchers are adopting BYOKG and GraphRAG to improve access to fragmented knowledge systems. These solutions aim to integrate data sources, enhancing efficiency in drug discovery.
Amazon Bedrock Offers AI-Powered Email Management for Public Sector
Amazon Bedrock introduces an AI-driven email management solution for public sector organizations to automatically classify and prioritize incoming communications. This system enhances responsiveness and optimizes resource allocation in local government settings where email volume is high.
Amazon QuickSight Introduces Multi-Dataset Topics and Relationships for Enhanced Analytics
Amazon QuickSight has launched Multi-Dataset Topics and Multi-Dataset Relationships, allowing analytical questions across multiple datasets without the need for pre-joining them into a single table. This new capability enables AI-generated SQL queries and supports seven distinct data modeling patterns, facilitating more efficient data analysis and management. The updates are part of efforts to streamline dataset management and enhance data governance.
Amazon Nova introduces selective unlearning for customizable content moderation
Amazon Nova has launched Customizable Content Moderation Settings (CCMS), allowing clients to selectively adjust model safeguards. This development is significant as it enables organizations to generate needed content while adhering to essential AI safety policies.
AWS Introduces Multi-Turn Reinforcement Learning in SageMaker AI for Enterprise Agents
AWS has introduced new capabilities for multi-turn reinforcement learning in Amazon SageMaker AI and Amazon Nova. These developments facilitate the training of AI agents that can handle complex workflows and multi-step tasks, improving their reliability and performance. This advancement marks a significant step in enhancing enterprise AI solutions such as support ticket resolution and content moderation.
Amazon Nova automates PII redaction in images for compliance
Amazon has introduced Nova, a foundation model that automates the redaction of PII in images. This tool addresses compliance challenges by ensuring sensitive information is accurately identified and redacted, reducing the risk of regulatory penalties for organizations sharing images containing potential PII.
Amazon Bedrock introduces tools to combat AI-generated phishing risks
Amazon Bedrock offers capabilities to detect and address AI-generated phishing, adapting to sophisticated attacks. This response is crucial as traditional phishing filters fail against today's contextually accurate threats.
HippoRAG Framework Integrates Amazon Services for Enhanced Knowledge Retrieval
HippoRAG is a new Retrieval Augmented Generation framework inspired by human memory systems. It utilizes Amazon Bedrock, Neptune, and personalized PageRank for efficient multi-hop reasoning and knowledge integration across documents.
BoltzGen Deploys on Amazon SageMaker AI for Protein Design
BoltzGen is now available on Amazon SageMaker AI, optimizing protein binder design processes by managing GPU infrastructure automatically. This integration allows researchers and developers to focus on design rather than infrastructure, significantly reducing operational overhead and costs in protein design projects.
Amazon Bedrock enhances resilience for LLM inference in production environments
Amazon Bedrock introduces resilience patterns to improve large language model (LLM) inference. These patterns focus on maintaining availability, response time, and cost-effectiveness, critical as generative AI scales in production.
Outpost VFX Achieves 8x Faster AI Training Using AWS
Outpost VFX improved their AI model training speeds by 8x by leveraging AWS infrastructure. This enhancement optimizes face replacement workflows, reducing project delays and improving overall efficiency in visual effects production.
IBS Software Develops Bilingual NER with Amazon Bedrock for Cargo Logistics
IBS Software created a bilingual Named Entity Recognition (NER) system utilizing Amazon Bedrock to process cargo logistics emails in English and Japanese. This solution achieves over 95% accuracy while reducing operational costs significantly, addressing challenges in manual interventions and accuracy trade-offs.
Amazon Nova models fine-tuned for better email data extraction accuracy
Parcel Perform has fine-tuned Amazon Nova models to improve data extraction accuracy from emails, achieving 94.77% accuracy. The optimization reduced inference latency and costs, enhancing e-commerce logistics operations.
Amazon Nova 2 Lite and Anthropic Claude streamline document digitization
Amazon Nova 2 Lite and Anthropic's Claude Sonnet 4.6 have been paired to create a cost-effective solution for digitizing scanned documents. This two-model pipeline reduces costs by two-thirds compared to single-model alternatives, providing high confidence name-to-face associations from yearbook scans.
PAR Technology Develops Secure Multi-Tenant LLM Analytics on AWS
PAR Technology has built a secure multi-tenant analytics system on AWS that uses row-level security. This solution allows business users to perform self-serve analytics without exposing sensitive data across tenants, addressing the complex challenge of data access and correctness.
Amazon QuickSight Introduces Hierarchy Filter for Simplified Dashboard Drill-Down
Amazon QuickSight has launched a new hierarchy filter feature, allowing dashboard authors to create multi-level filtering controls within a single interface. This update aims to reduce visual clutter and guide users through data exploration more efficiently by presenting data in a logical, hierarchical path.
uniopen customized Amazon Nova 2 Lite for retail moderation policies
uniopen, a digital communication platform, adapted Amazon Nova 2 Lite using supervised fine-tuning in Amazon SageMaker AI to meet its specific retail moderation policies. This customization allows uniopen to classify user interactions based on nine behavior categories and three subject types, which are unique to its business operations. The approach separates production moderation from correction, training, and deployment, using AWS services for a repeatable workflow.
Hugging Face Models Now One-Click Deployable to Amazon SageMaker Studio
AWS and Hugging Face have integrated deep-linking, allowing developers to move a model from Hugging Face directly into Amazon SageMaker Studio in a single click. This eliminates prior multi-step processes, enabling quicker model experimentation and deployment. The update is significant for faster AI development and deployment in enterprise environments.
Condé Nast implements AI-powered multimodal video discovery with Amazon Bedrock
Condé Nast partnered with AWS to develop an AI-powered multimodal video discovery solution using Amazon Bedrock and OpenSearch Service. This system reduces content discovery time from 250 minutes to under 2 minutes per task for its 140,000-video library. The solution addresses the inefficiency of manual video scrubbing and keyword-based searches by enabling intent-based semantic search across video transcripts, visuals, and audio.
Aderant Implements Intelligent Ticket Triage System with Amazon Nova Lite
Aderant, a legal software provider, developed an intelligent ticket triage system using Amazon Nova Lite through Amazon Bedrock. This system automates context gathering, classification, routing, and knowledge enrichment for support tickets, reducing manual effort and improving routing accuracy.
EXL Launches AI-Powered Medical IDP Solution on AWS to Reduce Claims Review Time
EXL introduced an AI-powered Medical Intelligent Document Processing (IDP) solution built on AWS to automate medical claims review. This solution combines IDP with domain-specific large language models to extract, summarize, and query medical information, aiming to reduce the over 100 minutes adjusters spend per case manually reviewing records.
AvioBook uses Amazon Bedrock AgentCore to develop turnaround insights from operational data
AvioBook, a Thales Group company, is using Amazon Bedrock AgentCore to build "Connected Analytics" for its AvioBook Connect platform. This integration aims to transform operational data from airline turnarounds into actionable insights, helping airlines reduce delays and associated costs.
University Startups develops Trinity, an AI solution for postsecondary planning for students with disabilities
University Startups developed Trinity, a conversational AI solution that assists students with disabilities in creating personalized postsecondary transition plans. Trinity helps students explore interests and goals to generate IDEA-aligned plans, and has expanded to over a dozen U.S. states with international expansion underway.
Panasonic Avionics uses agentic AI on AWS to accelerate IFEC diagnostics
Panasonic Avionics Corporation developed an agentic AI system with AWS to speed up diagnostics for its in-flight entertainment and connectivity (IFEC) systems. This solution, utilizing Amazon Bedrock, Amazon SageMaker, and AWS Glue, reduces the time needed to identify root causes of system issues across diverse aircraft configurations.
Pixieset integrates AI alt text generation using Amazon Bedrock, achieving 35% adoption
Pixieset, a photography business service, implemented an AI-powered alt text generator for images using Amazon Bedrock, reaching 35% adoption among applicable users within 16 months. This feature addresses the time-consuming task of manually writing alt text for photographers' portfolios, which often contain hundreds to thousands of images.
AWS Generative AI Innovation Center develops constraint programming system for NHL playoff scenarios
The AWS Generative AI Innovation Center created an automated system using constraint programming and custom tree search to determine NHL playoff clinching scenarios. This system provides a mathematically rigorous alternative to manual scenario generation, which has become time-consuming and error-prone due to complex tie-breaking rules.
AI as Accessibility Tool for Neurodivergent Professionals
AI is leveraged as an accessibility tool by neurodivergent individuals, improving executive function tasks. This usage highlights the gap in traditional productivity tools that often overlook neurodivergent needs.
NVIDIA Resiliency Extension (NVRx) integrates with PyTorch FSDP on Amazon EKS for fault-tolerant distributed training
NVIDIA Resiliency Extension (NVRx) has been integrated into PyTorch Fully Sharded Data Parallel (FSDP) training on Amazon EKS to address interruptions and idle time in large-scale distributed training jobs. This integration provides async checkpointing, in-process restart, and in-job restart capabilities, improving the efficiency and reliability of long-running GPU workloads.
Decathlon implements Chronos-2 for large-scale demand forecasting
Decathlon, a global sporting goods retailer, adopted Chronos-2 as a core component of its demand forecasting system to manage tens of thousands of products across multiple continents. This implementation aims to improve the accuracy and efficiency of predicting weekly sales for both short-term replenishment and long-term strategic planning. The company previously used a hybrid approach with Amazon SageMaker AI DeepAR and Holt-Winters, later incorporating Temporal Fusion Transformer (TFT).
GoDaddy Migrates Analytics Infrastructure to Amazon Quick, Improving Performance and Efficiency
GoDaddy migrated its business intelligence infrastructure from a legacy BI tool to Amazon Quick over two years. This migration reduced dashboard rendering times from over 15 minutes to under 5 seconds, cut dashboard count by 50%, and saved 15,000 hours annually, making AI-powered self-service analytics accessible to employees.
AWS details data preparation for supervised fine-tuning of large language models
AWS published a two-part series on preparing data for supervised fine-tuning (SFT) of large language models. The series covers foundational aspects like formatting, quality checks, and train/evaluation splits, as well as advanced strategies such as data subset selection, augmentation, and mixing. This guidance helps customize foundation models like Amazon Nova to meet specific production requirements by reshaping their behavior.
AI-powered metadata correction and harmonization workflow developed on AWS
A new workflow built on AWS uses AI for metadata correction and harmonization, addressing the challenge of standardizing large volumes of data. This system aims to automate the process of aligning metadata schemas and validating data integrity, which traditionally has been a manual bottleneck in data analysis.
Fanatics Betting and Gaming Implements Multi-Agent AI for Customer Support on AWS
Fanatics Betting and Gaming (FBG) developed a multi-agent AI customer support system on AWS to manage the complexities of sports betting inquiries, which vary by jurisdiction and surge during live events. This system aims to provide faster, more accurate customer service and reduce operational costs compared to human-only support models. The implementation addresses the challenge of diverse, time-sensitive customer queries in a highly regulated industry.
ONESTRUCTION develops Ishigaki-IDS foundation model for construction BIM workflows with AWS GenAIIC
ONESTRUCTION, a construction technology startup, developed Ishigaki-IDS, a foundation model specialized for Building Information Modeling (BIM) workflows in the construction industry, with technical advisory from AWS Generative AI Innovation Center (GenAIIC). This model addresses the challenge of data scarcity in specialized domains by using synthetic data generation and a three-stage training pipeline, making BIM attribute information review and management accessible to non-specialists.
Guide to Building a Serverless A2A Gateway for AI Agent Communication
A tutorial outlines building a serverless A2A gateway for efficient agent communication. It addresses operational burdens by centralizing agent routing and access control, which aids in managing multiple AI agents without complex integrations.
Amazon QuickSight Introduces Backup Strategy for BI Assets
Amazon QuickSight now allows the backup of BI assets using the AssetsAsBundle APIs to safeguard against data loss. Implementing this strategy is crucial for organizations relying on QuickSight for critical decisions, especially in regulated industries.
Implementing Multi-Environment Access for Claude Platform on AWS
This post details how to set up multi-environment access for Claude Platform on AWS (CPonAWS), covering authentication for AWS workloads, developer laptops, and external services. It outlines a three-account architecture to manage a single CPonAWS subscription across different environments while maintaining isolation.
Tata Elxsi develops IRIS platform on AWS for real-time industrial safety risk detection
Tata Elxsi built IRIS (Industrial Real-Time Intelligence System) on AWS to detect industrial safety risks in seconds using computer vision. This platform addresses challenges in traditional safety monitoring, such as reactive CCTV, human monitoring limits, and inconsistent policy enforcement, by providing continuous, automated detection.
BMW Group uses AWS-based system to detect cloud cost anomalies across 14,000 accounts
BMW Group developed Cloud Efficiency Analytics (CLEA), an in-house FinOps system on AWS, to monitor and detect cost anomalies across its 14,000 cloud accounts. This system forecasts expected cloud spend and alerts account owners daily when actual spending deviates from patterns, helping manage cloud costs effectively.