Dr. Karthik Vaidhyanathan along with Prof. Henry Muccini from the University of L’Aquila, Italy conducted a tutorial on Software Architecture in the Age of Agentic AI as part of the 19th European Conference on Software Architecture (ECSA, Core A-2025) held at Limassol, Cyprus from 15 to 19 September.
This tutorial brought together academic expertise and industry experience to provide a comprehensive view of Software Architecture in the Age of Agentic AI. The collaboration had both theoretical depth and practical applicability for conference attendees.
More details: https://agentic-sa.vercel.app/
Hiya Bhatt working with Dr. Karthik Vaidhyanathan presented a research paper on HarmonE: A Self-Adaptive Approach to Architecting Sustainable MLOps in the main research track of the 19th European Conference on Software Architecture (ECSA, Core A-2025) held at Limassol, Cyprus from 15 to 19 September. The work was a joint collaboration with Shaunak Biswas, Dr. Karthik Vaidhyanathan and Srinivasan Rakhunathan, Microsoft-India. Here is the summary of the paper as explained by the authors:
Machine learning powers much of what we use every day, but keeping the models that these systems use reliable and energy-efficient over time isn’t easy. Models drift, data changes, and retraining too often can drain resources. Retraining models too often wastes energy and resources, while doing too little can hurt performance.
HarmonE is our solution: an architectural approach that helps machine learning pipelines adapt on their own. It monitors accuracy, energy use, and data shifts to keep things running smoothly. It only takes action when needed, based on sustainability goals set in advance.
We tested HarmonE on a city’s traffic system using a digital twin. It adapted to new conditions while saving energy and maintaining performance. After all, sustainability isn’t just about quick fixes; it’s about building systems that can maintain their performance over time, even as conditions evolve.
Full paper: https://sa4s-serc.github.io/HarmonE/
Conference page: https://conf.researchr.org/home/ecsa-2025
September 2025