Novo Nordisk's manufacturing technology team developed an agentic AI system on AWS to manage complex documentation and retrieve critical manufacturing knowledge. By implementing a GraphRAG architecture that combines a knowledge graph with hybrid search, the team improved compliance visibility and significantly reduced the time required for subject-matter experts to answer process-related queries.
Key points
Using a GraphRAG architecture improves retrieval accuracy by grounding responses in a structured ontology rather than relying solely on vector search.
A clarify-before-retrieve agent reduces ambiguity and increases the precision of answers in complex domains.
Hybrid retrieval strategies that combine knowledge graphs with semantic search provide better traceability for regulated industries.
Iterative prototyping with subject-matter experts is essential to de-risk architecture and ensure the system meets user requirements.