Selected Work
AI Factory Reference Architecture
ProductionA complete validated reference design for hyperscale AI factory deployments — covering GPU cluster topology, RoCEv2 fabric design, NVMe-oF storage integration, and operational runbooks for 1,000+ GPU environments.
Lossless Ethernet Fabric for RDMA
PublishedEnd-to-end design and implementation of a 400G lossless Ethernet fabric optimized for RoCEv2 workloads. Includes PFC configuration, ECN tuning, DCQCN parameter optimization, and operational tooling for fabric health monitoring.
Post-Quantum Cryptography Migration Framework
ResearchA practical framework for migrating enterprise infrastructure to hybrid post-quantum encryption schemes, aligned with NIST PQC standards (CRYSTALS-Kyber, CRYSTALS-Dilithium, SPHINCS+). Includes threat modeling, migration playbooks, and testing methodology.
GPU Cluster Checkpoint Storage Design
PublishedTiered checkpoint storage architecture for 10,000-GPU training clusters. Covers failure probability analysis, checkpoint frequency optimization, fast restore design, and cost/performance tradeoffs across NVMe, NFS, and object storage tiers.