Avirup Saha

Across my career in AI research and applied machine learning, my work has focused on understanding how complex systems behave — whether we’re looking at cloud federations, large-scale IT observability pipelines, human information diffusion in social platforms, or enterprise workflows that evolve over time.

I earned my Ph.D. in Computer Science & Engineering from IIT Kharagpur, where my research centered on modeling complex event dynamics in temporal and networked systems. I developed methods for learning competitive temporal point processes and graph-based semi-supervised learning, contributing to a foundational understanding of how information, behaviors, and operational signals evolve over time. This work continues to inform how I design data-driven systems for enterprise decision-making.

I am currently part of the Research division at IBM, where I build systems and methods for predictive observability, AI-driven automation, and trustworthy decision systems. My work spans temporal point processes, multivariate time-series forecasting, graph-based semi-supervised learning, foundation models for data-centric applications, and agentic workflows for enterprise environments. I am driven by the idea that AI should not only model complex systems but anticipate how they change — and help organizations act with foresight rather than hindsight.

Over the years, I have published research in leading venues such as AAAI, SIGIR, NAACL, CIKM, IJCAI, INFOCOM, UAI, EMNLP, and IEEE Transactions. My contributions include advances in modeling business and IT dynamics, news and opinion diffusion, and cloud resource federation. I have also filed and received multiple U.S. patents, including work on KPI impact forecasting, entity explanation systems for data management, and temporal behavior modeling for anomalies.

Beyond research, I’ve led collaborations with product engineering and enterprise clients to translate research ideas into deployable solutions — particularly in business process observability, AIOps, and foundation models for enterprise automation. I find purpose in closing the gap between theory and practice, and in shaping systems that are robust, interpretable, and aligned with organizational goals.

Focus Areas:

• Predictive Observability & AIOps

• Enterprise Automation & Agentic AI

• Foundation Models for Data Platforms

List of publications

List of patents