Graph-Augmented LLMs for End-to-End Workflow Synthesis and Optimization

Li et al.’s work on integrating retrieval-augmented generation with knowledge graphs for supply chain management demonstrates the power of combining structured and unstructured data. This idea extends that principle, proposing a research direction where LLMs are tightly integrated with domain-specific knowledge graphs (representing data sources, processing steps, policies, and dependencies), enabling users to describe desired outcomes or processes in natural language. The system would then automatically synthesize end-to-end workflows, optimizing for efficiency, compliance, or resource usage. Unlike current approaches that focus on code generation for isolated tasks (e.g., Mani et al., 2023), this approach would allow for holistic, cross-system workflow design and real-time adaptation—potentially transforming how non-technical users interact with and manage complex data ecosystems.

References:

1. Enhancing Network Management Using Code Generated by Large Language Models. Sathiya Kumaran Mani, Yajie Zhou, Kevin Hsieh, Santiago Segarra, Trevor Eberl, Eliran Azulai, Ido Frizler, Ranveer Chandra, Srikanth Kandula (2023). ACM Workshop on Hot Topics in Networks.
2. Integrating Graph Retrieval-Augmented Generation With Large Language Models for Supplier Discovery. Yunqing Li, Hyunwoong Ko, Farhad Ameri (2024). Journal of Computing and Information Science in Engineering.