Abhinav U Sharma

Knowledge Graph Augmented Multi-Agent Systems

Abstract:

Knowledge Graph-Augmented Multi-Agent Systems offer a framework for improving how autonomous agents negotiate, coordinate, and maintain agreement-related workflows. The central idea is that agents can make better decisions when they operate over a shared semantic layer that captures entities, relationships, constraints, provenance, and agreement history. This talk examines how Knowledge Graphs can support Multi-Agent Systems in domains such as procurement, logistics, and supply-chain coordination, while also addressing the technical challenges created by shared-state architectures. Key issues include distributed transaction processing, semantic interoperability across heterogeneous ontologies, access control, temporal reasoning, and conflict resolution. The analysis highlights hybrid consistency models as a practical enterprise approach, using strong consistency for critical agreement terms and more flexible eventual consistency for secondary operational data.

Profile:

Abhinav Sharma is a Staff Machine Learning Engineer and a founding member of Docusign’s AI organization. In this role, Abhinav has led the architecture and scaling of the platform into an enterprise-grade system that processes 100M+ agreements annually for 25,000+ customer accounts across five global geographies. His contributions and research involve recommendation systems, asynchronous processing pipelines, vector storage systems, inference engines, chat systems, and semantic clustering systems for context-rich documents. 
His professional background includes a tenure at Microsoft, where he served as a Senior Software Engineer for Azure AI Search and Azure Networking organizations. At Microsoft, Abhinav was responsible driving critical initiatives for the multi-billion dollar U.S. DoD JEDI program, including strengthening the security and audit posture of the services using differential privacy, regional service buildouts, and the delivery of a zero-touch secret rotation system for a fleet of tens of thousands of machines. He also developed continuous end-to-end integration testing frameworks to detect potential large-scale outages across 50+ Azure regions.