Current

Microsoft Azure

Centralized Logging for CycleCloud Slurm: Simplify troubleshooting at scale with centralized log management for CycleCloud Workspace for Slurm. Published March 31, 2026.
- Code: slurm-log-collection, notes on configuring OpenTelemetry for Azure Monitor.
Running DeepSeek-R1 on NDv5 AMD MI300X VM: Run the DeepSeek-R1 reasoning model on a single Azure NDv5 MI300X VM. Published January 31, 2025.
- Contributed a performance fix to SGLang for DeepSeek on MI300X in virtualized environments.
Training large AI models on Azure: Train large AI models on Azure using CycleCloud and Slurm. Published March 28, 2023.
- Code: benchmark-opt175b.
ib-topo: InfiniBand topology discovery tool for Azure HPC clusters, now incorporated into AzHPC Images and AI Infrastructure on Azure.
GB200: Getting NDv6 GB200 nodes running for AI training.
- Notes on using GB200 (NDv6) nodes: topology, NVLink/IMEX configuration, and Slurm.
- Notes on configuring NVMe for GB200.
- Contributed a fix to TorchTitan for Flux image-generation benchmarking.

Detect and Classify Grouper Sounds: Detect and classify fish sounds from passive acoustic recordings using CNNs.
Bering Sea Drift and Spill Vessel: Estimate oil spill risk to Alaskan shores from gounded vessels.
S2S - Sea Ice Guidance: Provide statistically generated probabilistic sea ice guidance for the three-week to seasonal time scale.
Coral Spawning Detector: Develop CNN based detector to identify spawning coral in aquaria.
Arctic Oil Spill Risk Assessment: Provide estimates of risk to areas used for subsistance gathering from small “operational” oil spills.
IFCB: Develop classifiers and data pipelines for IFCB instrumentation to classify and derived statistics about observed harmful algal bloom species of interest.

Developed, optimized, and applied a distributed, parallel numerical model validated by in-situ observations to study coastal and estuarine physics, aquatic light environments, and sediment dynamics as part of my PhD work.