Category: Digital Transformation

Bayer Crop Science has deployed a new Agentic Decision Engine (ADE) designed to simplify the complex data requirements of its global plant breeding operations. The system serves as a natural language interface over the company’s existing Connected Decision Engine (CDE), allowing researchers, breeders, and data scientists to interact with analytics workflows without manually navigating fragmented operational logs, dashboards, and metadata stores.

Solving operational complexity

Previously, Bayer’s breeding research—which integrates genomics, field trials, and environmental imagery—relied on hundreds of machine learning configurations. As adoption increased, the challenge shifted from executing models to understanding the results. Users faced significant delays when investigating workflow failures or tracing data origins. The ADE addresses this by employing a multi-agent architecture where a supervisor agent routes queries to specialized agents, such as a ‘Troubleshooting Agent’ or ‘Domain Expert Agent,’ which retrieve necessary context and provide answers directly.

According to the company, the engine is resolving approximately 80% of user inquiries without requiring engineer intervention. For instance, diagnostic tasks that once required over an hour of manual log analysis can now be completed in under 10 minutes.

Technical implementation and architecture

The ADE is powered by Amazon Bedrock AgentCore, which provides a framework-agnostic runtime for managing agent sessions and tool access. The architecture includes four distinct layers: a user interface, agent orchestration via LangGraph, specialized domain agents, and a data plane utilizing Amazon DynamoDB and Amazon Aurora. To ensure secure and scalable operations, Bayer integrated AgentCore Gateway to manage authentication and authorization independently from agent code, preventing the need to share tokens across trust boundaries.

The development process involved close collaboration with AWS Unified Operations. Technical Account Managers and generative AI specialists participated directly in the design phase, conducting reviews that targeted specific failure modes and capacity risks. This engagement helped the team transition the system from a prototype to a production-ready environment, incorporating security measures such as role-based access control and structured output validation.

Future directions

Since the rollout, over 100 users across Bayer Crop Science have adopted the platform. The company is currently exploring ways to enable interoperability between agents running in different environments, aiming to allow hybrid cloud communication without manual configuration. By adding an intelligent layer atop its existing analytics infrastructure, Bayer intends to accelerate scientific discovery while maintaining rigorous control over its ML-driven research pipelines.

Source: aws.amazon.com
Original article: https://aws.amazon.com/blogs/machine-learning/how-bayer-crop-science-built-an-agentic-decision-engine-on-amazon-bedrock-agentcore-with-aws-unified-operations/

IskayPet, a major pet care company operating in Spain, Portugal, and France, has successfully deployed an internal knowledge management assistant to address operational inefficiencies. By leveraging Amazon Bedrock Managed Knowledge Bases, the company has unified documents previously scattered across hundreds of Microsoft SharePoint files, reducing the time required to retrieve information from several minutes to seconds.

Overcoming Fragmented Data

Prior to the deployment, IskayPet employees—spanning retail, veterinary, and procurement operations—relied on manual searches through documents stored in different languages and locations. This process was prone to inaccuracies and grew increasingly difficult as the company expanded its physical footprint, which now includes over 330 stores and 4,500 employees. To modernize this workflow, the internal team sought a solution capable of providing grounded, natural-language answers based on existing documentation without necessitating a massive data migration.

Implementation with Managed RAG

The company utilized Amazon Bedrock Managed Knowledge Bases to implement a Retrieval Augmented Generation (RAG) system. This approach allowed the team to bypass the complexities of managing vector databases, embedding pipelines, and ingestion infrastructure. The solution connects directly to IskayPet’s existing SharePoint repositories, utilizing advanced indexing to process text and embedded images from documents such as PDFs and slide decks.

A critical feature of the system is its document-level access control, which preserves existing SharePoint permissions. The assistant only returns passages that a specific employee is authorized to view, ensuring data governance is maintained without rebuilding authorization models. Furthermore, the architecture employs hybrid search, combining semantic and keyword-based retrieval to ensure precise results, such as product codes, are captured effectively.

Development and Scalability

The development of both the frontend and the underlying AWS architecture was accelerated by using Kiro, an agentic development environment, running on Anthropic’s Claude Haiku 4.5 model. This allowed a small team to build and iterate on the assistant rapidly. By structuring the system with a dedicated knowledge base per assistant, IskayPet can easily scale to new business areas or countries by pointing the system to the relevant document libraries.

According to the company, the assistant has already driven significant productivity gains, with search times dropping from 3–5 minutes to approximately 7 seconds. As the company continues to refine the system, it plans to integrate Amazon Bedrock Guardrails to enforce responsible AI policies and evaluate new foundation models as they become available on the platform.

Source: aws.amazon.com
Original article: https://aws.amazon.com/blogs/machine-learning/how-iskaypet-unified-corporate-knowledge-with-amazon-bedrock-managed-knowledge-bases/

U.S. Hunger, a nonprofit organization dedicated to addressing food insecurity, is leveraging Amazon Quick to centralize its operational and financial data. The initiative aims to support the organization’s mission by providing leaders with direct access to information, facilitating faster decision-making for its food-assistance program, Full Cart.

Data-Driven Mission Scaling

With a history of delivering over 250 million meals, U.S. Hunger has experienced 500 percent growth over four years with only minimal headcount expansion. To maintain this pace, the organization integrated Amazon Quick to pull data from SQL Server into the SPICE in-memory engine. This setup allows CEO Rick Whitted and other leaders to explore core datasets, including survey responses and financial performance indicators, without requiring constant custom data pulls from the technology team.

Centralized Collaboration via Quick Spaces

The implementation of “Quick Spaces” has served as a central hub for the leadership team. By establishing shared definitions for key performance indicators (KPIs) and providing access to financial documents and planning tools, the organization has improved coordination across departments such as finance, technology, and program management. This approach allows staff to align on business logic and performance goals, reducing internal friction in planning and analysis.

Social Impact and Future Plans

Beyond internal operations, U.S. Hunger is analyzing data to understand the complex pressures behind food insecurity, such as income, housing, and health. The organization utilizes classification codes, known as Z codes, to visualize these social determinants of health. These insights help leadership identify early warning signals of food instability, enabling more proactive intervention.

Looking ahead, U.S. Hunger plans to expand its data capabilities by building an enterprise KPI framework. This upcoming infrastructure will integrate data from diverse systems—including NetSuite, Salesforce, and Impact Manager—into Amazon Redshift. The organization also envisions transforming its internal data resources into a broader, publicly accessible platform to support external research on hunger, while simultaneously exploring enterprise offerings to help sustain its mission.

Reflecting on the integration of AI tools, Rick Whitted emphasized that the primary goal is not merely to automate, but to bridge the gap between complex data and the human-centric needs of the families served. “This is not about bringing humans into the AI loop. It’s about bringing AI into the human loop,” he noted.

Source: aws.amazon.com
Original article: https://aws.amazon.com/blogs/machine-learning/how-u-s-hunger-connects-leaders-and-data-using-amazon-quick/

Microsoft has released version 0.102 of its PowerToys utility suite, introducing new features designed to enhance mouse management for users. The update adds two specific tools: CursorHide and Button Lock.

In addition to the new mouse utilities, the latest release includes updates to the application’s Command Palette and introduces new command-line tools. These additions are part of Microsoft’s ongoing development of the open-source suite, which provides system utilities intended for power users to tune and streamline their Windows experience.

The integration of CursorHide and Button Lock further expands the toolkit’s capabilities, following previous updates to the platform’s interface and utility set. Users can access these features by updating to the latest version of the PowerToys software.

Source: BleepingComputer
Original article: https://www.bleepingcomputer.com/news/microsoft/microsoft-powertoys-adds-cursorhide-button-lock-mouse-tools/