Jayaram K Radhakrishnan

I am a senior technical leader at IBM Research, working on agents and automation. My current focus is agentic memory — how autonomous agents form, store, retrieve, and reuse what they learn from experience.

I publish as K. R. Jayaram; my patents are filed under my full legal name, Jayaram Kallapalayam Radhakrishnan.

Agents that run real work do not fail because a single model call is weak. They fail because nothing carries forward: the same subtask gets re-derived, hard-won context is dropped between runs, and useful trajectories are thrown away instead of becoming reusable knowledge. Memory is the systems problem underneath that, and it is the one I spend my time on.

What I work on

Agentic memory. Turning agent trajectories into durable, retrievable knowledge — generating memory from execution traces, generalizing subtask-level knowledge so it transfers beyond the run that produced it, deciding what is worth keeping at all, and packaging memory so it survives deployment.

Agents and automation. Conversational generation of enterprise workflows, in-context example selection and ordering for reliable API sequence generation, and cost-aware multimodal agents that spend inference budget where it actually buys accuracy.

Federated and distributed learning. Private and decentralized aggregation, intelligent participant selection, and the security properties that make cross-silo learning practical.

Systems for large-scale ML. Multi-tenant deep learning platforms, elastic scaling of training workloads, and scheduling that is aware of what the job is actually doing.

The through-line across two decades — from publish/subscribe and event correlation, through cloud elasticity and deep learning infrastructure, to federated learning and now agents — is the same: building the distributed systems substrate that makes a new class of computation dependable enough to run in production.

Selected publications

• VOILA: Value-of-Information Guided Fidelity Selection for Cost-Aware Multimodal Question Answering Rahul Atul Bhope, K. R. Jayaram, Vinod Muthusamy, Ritesh Kumar, Vatche Isahagian, Nalini Venkatasubramanian arXiv, 2026·arXiv
• Trajectory-Informed Memory Generation for Self-Improving Agent Systems Gaodan Fang, Vatche Isahagian, K. R. Jayaram, Ritesh Kumar, Vinod Muthusamy, Punleuk Oum, Gegi Thomas arXiv, 2026·arXiv
• OptiSeq: Ordering Examples On-The-Fly for In-Context Learning Rahul Atul Bhope, Praveen Venkateswaran, K. R. Jayaram, Vatche Isahagian, Vinod Muthusamy, Nalini Venkatasubramanian EMNLP Findings, 2025·arXiv·DOI
• FLOW-BENCH: Towards Conversational Generation of Enterprise Workflows Evelyn Duesterwald, Siyu Huo, Vatche Isahagian, K. R. Jayaram, Ritesh Kumar, Vinod Muthusamy, Punleuk Oum, Debashish Saha, Gegi Thomas, Praveen Venkateswaran EMNLP Industry Track, 2025·arXiv·DOI
• FLIPS: Federated Learning using Intelligent Participant Selection Rahul Atul Bhope, K. R. Jayaram, Nalini Venkatasubramanian, Ashish Verma, Gegi Thomas Middleware, 2023·arXiv·DOI
• FfDL: A Flexible Multi-tenant Deep Learning Platform K. R. Jayaram, Vinod Muthusamy, Parijat Dube, Vatche Ishakian, Chen Wang, Benjamin Herta, Scott Boag, Diana Arroyo, Asser N. Tantawi, Archit Verma, Falk Pollok, Rania Khalaf Middleware, 2019·arXiv·DOI

All publications →

Patents

10 granted inventions (18 issued patents worldwide) and 8 pending applications, spanning agentic memory, AI agents, federated learning, and deep learning infrastructure.

All patents →

Elsewhere

My IBM Research page is the official one. Find me on LinkedIn — that is the best way to reach me. Publication records: Google Scholar · DBLP.