My research investigates semantic memory as a security-critical infrastructure component for autonomous and multi-agent AI systems. The central thesis is dual: semantic caching introduces a new class of attack surfaces — cache poisoning, prompt-injection amplification, cross-tenant leakage, similarity-boundary manipulation — and, when designed with security intent, simultaneously becomes a more capable defensive layer than any control operating outside the agent–LLM path. The same equivalence-driven control-plane pattern transfers across non-LLM domains, including semantic circuit caching for hybrid quantum–classical workloads and vision-grounded agentic perception in closed-loop simulation environments.

This research is grounded in a long engineering practice at the intersection of agentic systems, large-scale data platforms, and real-time infrastructure: declarative system state, policy, and intent expressed as code through custom resource models and control planes; near-real-time telemetry pipelines processing multi-terabyte volumes across heterogeneous environments; and AIOps, MLOps, DevSecOps, and Site Reliability Engineering practices applied so the resulting systems stay observable, explainable, and operable at scale.

That practice comes from architecting and operating distributed systems in high-reliability and regulated domains: reactive and fault-tolerant platforms in banking and finance, GxP-aligned Manufacturing Execution Systems for pharmaceutical and biotech environments, and industrial automation and robotics for semiconductor manufacturing, including FOUP transport and SECS/HSMS integrations.

Research interests · semantic memory security · agentic control planes · approximate-retrieval governance · lifecycle FSMs and pre-execution refusal · multi-agent observability and audit substrates

Selected work · Semantic Equivalence as a Control-Plane Primitive: Improving Reliability of Distributed Agentic Communication Systems, IEEE GLOBECOM 2026 (under review); Agentic Data Services: A Control-Plane Architecture for Adaptive Data Workflows, IEEE BigDataService 2026 (under review); Agentic Control Planes for Hybrid Quantum–Classical Systems: A Kubernetes-Native Approach, IEEE QCE 2026 (under review); Policy-Aware Supervisory Control for Observable Embodied Biomedical Robotic Systems, IEEE CBS 2026 (under review); plus accepted contributions on agentic-dataset lifecycle FSMs and decentralised pre-execution refusal at CCECE 2026 and ANNSIM 2026.

LinkedIn