Researcher and engineer working on agentic AI, multi-agent LLM systems, and distributed systems infrastructure. I am especially interested in optimistic concurrency control for shared LLM state, formal verification of multi-agent coordination protocols, and the engineering of reliable autonomous AI infrastructure.

S-Bus: Automatic Read-Set Reconstruction for Multi-Agent LLM State Coordination. A Rust transactional middleware that uses a server-side DeliveryLog to automatically reconstruct each agent's read-set from observed HTTP traffic, enabling optimistic concurrency control over shared LLM state without agent-side SDK changes. Empirical safety parity demonstrated against PostgreSQL 17 SERIALIZABLE and Redis 7 WATCH/MULTI across 884,110 commit attempts on three independent backends. Mechanised in TLAPS, TLC, and Dafny.

• System: sbus — Rust workspace (server, baselines, proxy)
• Mechanised proofs: sbus-formals — TLA+, TLAPS, Dafny
• Experiments: sbus-experiments — Python harness reproducing every paper measurement

Preprint: 2026 (arXiv link — TBA).

Eight years of software-engineering experience across distributed backends, production AI systems (RAG over compliance content, and full-stack applications, in addition to the research above.