An Empirical Evaluation of Machine Learning for Hardening Security Devices in Data Networks

Evaluated machine learning approaches to improve robustness of security devices in networked environments, combining traffic analysis, anomaly detection, and deployment tradeoffs.

Understanding Failure Modes in Multi-Agent LLM Systems (Working)

Exploratory work on when multi-agent setups actually help vs hurt: compounding errors, coordination overhead, and how to design evaluation that detects silent failures.

Draft in progress — will be linked here and on Google Scholar when available.

Probing RAG Pipelines via Perturbation-Based Evaluation (Planned)

A planned line of work on evaluating retrieval-augmented generation by perturbing both documents and queries, and scoring the sensitivity and robustness of systems end-to-end.