Live
Cloudflare Images Analytics Dashboard Adds 30‑Day Delivery VisibilityAmazon ECS Adds Native Blue/Green, Canary, and Linear Deployments for Safer Container RolloutsGemini agentic tools add LLM‑driven automation to Partner Marketing StudioGitHub Actions makes RHEL 9/10 runner images generally available for larger workloadsIntegrating Decider Models to Reduce Latency and Cost in AI Agent WorkflowsAWS Emulator Enables Local Testing of 200+ Services, but Lacks IAM ValidationSageMaker AI monitoring adds automatic latency drift, idle GPU, and configuration gap detection for inference endpointsEmbedding Proactive Governance into Enterprise‑Scale Agentic AICloudflare Images Analytics Dashboard Adds 30‑Day Delivery VisibilityAmazon ECS Adds Native Blue/Green, Canary, and Linear Deployments for Safer Container RolloutsGemini agentic tools add LLM‑driven automation to Partner Marketing StudioGitHub Actions makes RHEL 9/10 runner images generally available for larger workloadsIntegrating Decider Models to Reduce Latency and Cost in AI Agent WorkflowsAWS Emulator Enables Local Testing of 200+ Services, but Lacks IAM ValidationSageMaker AI monitoring adds automatic latency drift, idle GPU, and configuration gap detection for inference endpointsEmbedding Proactive Governance into Enterprise‑Scale Agentic AI
AWS

GraphRAG architecture with Bedrock AgentCore streamlines pharma document retrieval

AI SummaryPowered by AI

Novo Nordisk replaced manual spreadsheet tracking with an AI‑driven GraphRAG system built on Amazon Neptune, OpenSearch Serverless, and Bedrock AgentCore. The change cuts document‑lookup time from hours or days to minutes, directly impacting compliance and operational efficiency for engineers.

Novo Nordisk migrated its manufacturing process documentation workflow from fragmented spreadsheets to a managed AI stack that combines a knowledge graph, hybrid search, and Bedrock AgentCore agents. Practitioners care because the new stack delivers minute‑scale answers while preserving version control, scope filtering, and traceability required in regulated environments.

The limitations of keyword and vector search

Traditional keyword search returns raw documents without awareness of document lifecycle state, site‑specific scope, or inter‑document relationships. A pure vector‑based Retrieval‑Augmented Generation (RAG) improves recall for natural‑language queries but still struggles with exact identifiers, multi‑hop relationship queries, and, crucially, indicating when no relevant content exists. In a pharmaceutical setting, those gaps translate to compliance risk and wasted engineering effort.

GraphRAG architecture details

The solution adopts a GraphRAG pattern: a knowledge graph stored in Amazon Neptune models document metadata, versioning, and explicit relationships (e.g., which SOP references which equipment). Parallel to this, a hybrid search index in Amazon OpenSearch Serverless provides both semantic vector search and exact‑match lookup over document contents. The two stores remain separate and are joined at query time, allowing each engine to specialize without compromising performance.

Agent orchestration with Bedrock AgentCore

Business logic lives in agents built with the Strands Agents SDK and executed on Amazon Bedrock AgentCore. AgentCore supplies a managed runtime that isolates sessions, orchestrates tool calls to both Neptune and OpenSearch, and integrates with AWS IAM for fine‑grained access control. A built‑in “guided mode” prompts users to clarify ambiguous queries before any retrieval occurs, improving answer precision and auditability.

Operational implications

From a cloud‑platform perspective, teams must provision and monitor two managed services (Neptune and OpenSearch Serverless) and ensure their IAM policies align with the regulated data handling requirements. The agent runtime adds observability hooks that surface latency, error rates, and usage patterns, enabling SREs to set alerts and auto‑scale based on query volume. Security engineers should verify that AgentCore sessions inherit the least‑privilege IAM roles and that all data accesses are logged for compliance audits.

Related CloudNinjas coverage: AWS.

What This Means For Practitioners

AI engineers should evaluate GraphRAG when document provenance and relationship queries are core requirements, and design agents that explicitly mediate between graph and vector stores. Cloud engineers need to plan for separate service scaling and IAM boundaries, while DevOps/SRE teams must instrument AgentCore metrics for reliable operation. Security teams should treat the managed AgentCore runtime as a trusted execution environment, but still enforce strict role‑based access and audit trails for any document retrieval.

Originally published atAWS Machine Learning Blog