Position: Collaborative Agentic AI Needs Interoperability Across Ecosystems
International Conference on Machine Learning (ICML), Position Paper Track · 2026
About Me
I am an AI Security Researcher at the Microsoft Security Response Center (MSRC) in Cambridge, UK, working on AI Security Research and Vulnerability Response, with a focus on Agentic Security.
Previously: I completed my Ph.D. in Computer Science at the Scalable Computing Systems Lab, EPFL, under the supervision of Prof. Anne-Marie Kermarrec, where I designed efficient and privacy-preserving systems for Decentralized and Federated Learning. During my Ph.D., I interned in the Azure Office of the CTO at Microsoft Research Cambridge, where I worked on deterministic security guardrails for autonomous AI agents, and visited the Camera Culture Group at MIT Media Lab under Prof. Ramesh Raskar, where I worked on collaborative AI in open decentralized networks and continued to collaborate closely with his lab.
Before that, I completed my bachelor’s degree in Computer Science and Engineering at the Indian Institute of Technology Mandi, where I received the President of India Gold Medal and worked with Dr. Manas Thakur on Automating Loop Parallelization for TornadoVM as part of my bachelor’s thesis.
Research Interests
AI SecurityAgentic SecurityDistributed SystemsArtificial IntelligenceMachine LearningDecentralized LearningPrivacy and Security
Education
- Ph.D. in Computer Science2021 – 2026École Polytechnique Fédérale de Lausanne (EPFL)Worked on making decentralized AI systems practical, efficient, and reliable.Read Thesis →
- Visiting Ph.D. in Computer Science2024 – 2025Massachusetts Institute of Technology (MIT)Designed and investigated infrastructure for autonomous collaborative agentic AI.
- B.Tech. in Computer Science and Engineering2017 – 2021Indian Institute of Technology (IIT) MandiGPA: 9.38/10.0
- Exchange Student in Computer Science2019 – 2020RWTH Aachen UniversityGPA: 1.0/1.0
Updates
🎓 Successfully defended my PhD!
Officially a Doctor of Philosophy in Computer Science from EPFL.
🎉 Our paper "Communication-Efficient Secure Aggregation in Decentralized Learning" (CESAR) has been accepted to SRDS 2026.
The first decentralized protocol to achieve both privacy and communication efficiency through secure aggregation.
🎉 Our position paper "Collaborative Agentic AI Needs Interoperability Across Ecosystems" has been accepted to the ICML 2026 Position Paper Track.
Arguing for minimal interoperability standards before the agentic AI ecosystem fragments.
🎉 Our paper "Orbit: Efficient Agentic Inference using Priority Scheduling" has been accepted to EuroMLSys 2026.
A trajectory-aware scheduler that deprioritizes agentic tasks predicted to fail.
🚀 Joining the Microsoft Security Response Center (MSRC) as an AI Security Researcher.
Moving to Cambridge, UK to work on AI Security Research and Vulnerability Response, with a focus on Agentic Security.
Selected Publications
A selected list. See the full list of publications or my Google Scholar.
Optimizing Agent Planning for Security and Autonomy
14th International Conference on Learning Representations (ICLR) · 2026
Boosting Asynchronous Decentralized Learning with Model Fragmentation
Proceedings of the ACM Web Conference (WWW) · 2025
Epidemic Learning: Boosting Decentralized Learning with Randomized Communication
37th Annual Conference on Neural Information Processing Systems (NeurIPS) · 2023
Recent Publications
Communication-Efficient Secure Aggregation in Decentralized Learning
International Symposium on Reliable Distributed Systems (SRDS) · 2026
Orbit: Efficient Agentic Inference using Priority Scheduling
Proceedings of the Sixth European Workshop on Machine Learning and Systems (EuroMLSys) · 2026
SONAR: Benchmarking Topology and Collaboration in Decentralized Learning
8th Conference on Machine Learning and Systems (MLSys) · 2026
HarMoEny: Efficient Inference of MoE Models
IEEE International Parallel and Distributed Processing Symposium (IPDPS) · 2026
SONAR Web: A Platform-Agnostic Framework for Real-Time Decentralized Learning Across Heterogeneous Edge Clients
Proceedings of the Federated Learning and Edge AI for Privacy and Mobility (FL-EdgeAI) · 2025
Invited Talks
Jul 31, 2026
Security Vulnerability Response for AI: Building Secure and Autonomous Agents
Invited talk @ Microsoft x CyberFirst Summer Symposium Day.
Mar 19, 2025
Decentralized Systems for Machine Learning
Invited talk @ MIT Media Lab.
SlidesMar 11, 2025
Decentralized Systems for Machine Learning
Invited talk @ McGill University.
SlidesMay 4, 2021
Scalability of Prolog: Real-world logic applications
Lecture @ IIT Mandi.
SlidesFeb 7, 2020
Adversarial training for blind-spot removal
HiWi talk @ RWTH Aachen University.
SlidesAcademic Service
- 2026: ICML’26 (Silver Reviewer); NeurIPS’26; AISec’26; TMC; AgenticOS; CODEC-FM;
- 2025: TMLR; NeurIPS’25; CFAgentic@ICML’25;
- 2024: NeurIPS’24 (Top Reviewer);
- Assistant Reviewer: OSDI’24; NETYS’24; SRDS’23;