SMART WORLDJOURNALSubmit Manuscript
← All technology news

Advanced Manufacturing

Novo Nordisk Deploys AI-Powered AgentCore to Streamline Manufacturing Documentation

Novo Nordisk leverages Amazon Bedrock AgentCore and GraphRAG architecture to transform its global pharmaceutical manufacturing documentation management, reducing search times from days to minutes and enhancing compliance in a highly regulated environment.

NEWS

Novo Nordisk, a leading pharmaceutical company specializing in treatments for chronic diseases, has modernized its manufacturing process document management using an AI-driven intelligent support system built on Amazon Bedrock AgentCore. This system addresses challenges managing thousands of interrelated controlled documents critical for product quality and regulatory compliance across global production sites.

The complexity arises from frequent document updates due to regulatory changes, process improvements, equipment modifications, and audit findings, demanding precise version control, scope alignment, and approval workflows. Traditional manual tracking with spreadsheets and fragmented tools proved time-consuming, error-prone, and difficult to scale, risking outdated information circulation.

To overcome this, Novo Nordisk implemented a GraphRAG solution combining a curated knowledge graph in Amazon Neptune with hybrid semantic and exact-match search via Amazon OpenSearch Serverless. This architecture treats document structure, relationships, and lifecycle states as first-class entities, enabling precise retrieval and connection tracing within the vast document corpus.

The AI assistance is powered by agent frameworks running on Amazon Bedrock AgentCore, orchestrating query understanding, guided disambiguation, hybrid retrieval, graph expansion, and a reranking stage, all backed by Anthropic Claude models. A 'guided mode' clarifies ambiguous queries through graph-based interactive prompts before searching, reducing the risk of out-of-scope or hallucinated answers.

Documents populate the system through a serverless ingestion pipeline with multimodal processing automated by AWS Step Functions, Lambda, and Bedrock Data Automation. The platform enforces grounding exclusively in approved, effective documents, explicitly flags missing or ambiguous content, and enables traceable, referenced responses vital for regulatory audits.

By converting a vault of roughly four million documents into an intelligent, answer-focused search engine guided by a fixed ontology reflecting real operational taxonomies, Novo Nordisk dramatically reduces subject matter expert task times from hours or days to minutes. This improvement enhances operational agility and ensures higher confidence in compliance-critical information.

The system’s design emphasizes transparency, auditability, and scalability suitable for the regulated pharmaceutical domain, with automated evaluation harnesses measuring retrieval relevance, version validity, scope accuracy, and citation fidelity. This foundation also supports future expansion into agentic workflows for document generation and impact analysis without compromising rigor.

This collaborative innovation, developed jointly by Novo Nordisk and AWS with iterative prototyping involving domain experts, exemplifies how AI and structured knowledge integration can revolutionize enterprise manufacturing documentation management, ultimately accelerating the delivery of life-saving medicines while maintaining stringent quality standards.

Source & further reading

Novo Nordisk modernizes manufacturing process documents with Amazon Bedrock AgentCore ↗

Source published 2026-11-16. Claims and limitations should be checked against the original announcement or research.