Research
My research sits at the intersection of AI reasoning, decentralized systems, and software security — currently focused on making LLM-based agents safe enough for high-stakes environments like finance and security operations.
How do we make LLM-based agents trustworthy? Multi-agent architectures for code and test generation (HPCAgentTester, Smartify, UniTranslator), hallucination and rationality analysis, benchmark integrity for LLM serving (vLLM on OpenXLA), and AI safety for security-critical workflows — including LogInject, our USENIX Security 26 study of prompt injection in SOC log analysis.
Decentralized infrastructure that doesn't trust any single party: TEE coordination across heterogeneous hardware (DHTee), decentralized FaaS, smart contract vulnerability detection and repair across Solidity and Move (Smartify — ACM SIGSOFT Distinguished Paper), privacy-preserving transactions and fair exchange, and on-chain AI inference.
Where it started: optimizing Web Virtual Reality (Rice University thesis), WebXR and A-Frame, decentralized social VR in the browser, and speech interfaces — work that became a decade of community teaching through Mozilla and Google Developer programs.
Quantum contrastive word embeddings with variational circuits for near-term quantum devices (QuCoWE) — exploring what NISQ-era hardware can realistically contribute to representation learning.