Research Publications
📄 SecAssure · 2026
DelegateGuard: Purpose-Bound Capabilities for Delegated LLM Agent Workflows
A runtime reference monitor that converts a user's authorization into attenuable, purpose-bound capabilities for delegated LLM agents, mediating every data-moving action outside the model and cascading revocation through the delegation graph. It blocks exfiltration, authority amplification, and revocation-replay attacks, and cut multi-turn data leakage from 0.844 to 0.20 with full task completion at microsecond per-decision overhead.
Read on arXiv📄 ICSA · 2026
Exploring the Reasoning Depth of Small Language Models in Software Architecture: A Multidimensional Evaluation Framework Towards Software Engineering 2.0
A multidimensional framework benchmarking 10 state-of-the-art Small Language Models (SLMs) on their reasoning capabilities for Architectural Decision Records generation, analyzing the trade-offs between model size, context windows, and performance.
Read on IEEE Xplore