Mr Mahesh Krishnan1
1Fujitsu Australia, Melbourne, Australia
Biography:
Mahesh Krishnan is the CTO for Fujitsu in Oceania, where his main role is to influence, and drive sustainable digital transformation using the key technologies including AI and Computing (HPC, Quantum, etc.).
He has previously been CTO at companies in the Energy and Health domain. He has written a couple of technical books and is the recipient of Microsoft’s Most Valuable Professional award for several years.
Abstract:
Background
Retrieval-Augmented Generation (RAG) is an effective technique for grounding large language models (LLMs) in enterprise knowledge by retrieving and incorporating relevant external content into the generation process. However, traditional RAG approaches often rely on shallow, keyword-based retrieval and are constrained by limited context windows. These limitations reduce their effectiveness when applied to large-scale, complex, and heterogeneous data environments typical of modern enterprises.
Method
To overcome these challenges, we introduce Knowledge Graph-enhanced Retrieval-Augmented Generation (KG-enhanced RAG), a novel method that integrates structured, semantically rich knowledge graphs into the RAG pipeline. Knowledge graphs represent entities, relationships, and hierarchies as interconnected nodes and edges, supporting semantic retrieval, entity disambiguation, contextual grounding, and enhanced explainability. This approach includes the automated construction of knowledge graphs from diverse enterprise data sources such as regulations, technical manuals, legal documents, and video transcripts.
Results
KG-enhanced RAG significantly expands the referable token scope – from millions to over ten million – without requiring retraining of the underlying language model. The result is a more scalable, accurate, and interpretable generative AI system capable of logical reasoning and providing transparent, traceable outputs.
Conclusion
Knowledge Graph-enhanced RAG offers a robust and scalable solution for enterprise-grade generative AI. It enables more intelligent access to complex knowledge, increases transparency, and supports a wide range of real-world applications across industries. This presentation provides an overview of the technology and demonstrates its application across multiple industry use cases.
