02Work

Work

Systems and projects I've led or built, and the open-source code behind some of them.

Fig. 1ML pipelineMove the pointer along the pipeline

Data, train, eval, and serve, with a feedback loop from serving back to data.

Selected work

Kempner AI cluster
Architected the institute AI cluster, which has grown to 1,100+ data-center GPUs. It ranked third among US academic supercomputers on the TOP500 in November 2024 and again in June 2026.
A100H100H200InfiniBandSlurm
KempnerInsight
A web app that monitors the AI cluster in real time: jobs by user and lab, GPU and node health, and utilization across compute, memory, power, and InfiniBand.
AWSDashboards
Kempner Computing Handbook
The guide to using the Kempner AI cluster, accessed from 110+ countries.
Documentation
StarGlass
Directed the engineering and launch of an AWS platform that hosts 500+ TB of sky data for researchers worldwide.
AWSData platform
PRIMECare
An online medical decision-support platform, hosted on AWS São Paulo and deployed in Argentina.
AWSHealth

KempnerForge
PyTorch-native framework for fault-tolerant, large-scale distributed training of foundation models on AI clusters.
KempnerInstitute/KempnerForge21 starsUpdated Sep 2026PyTorchFSDPCUDASlurm
KempnerPulse
Real-time GPU monitoring dashboard for DCGM metrics, showing accelerator efficiency across the cluster.
KempnerInstitute/kempnerpulse40 starsUpdated Aug 2026GPU telemetryDCGM
hpc-agentic-recipes
Recipes for deploying open-weight models on AI clusters with vLLM and SGLang.
KempnerInstitute/hpc-agentic-recipes14 starsUpdated Aug 2026vLLMSGLangSlurm