twh@waynehendricks.com · waynehendricks.com · github.com/twh · ORCID 0000-0001-9111-8968
Paige.AI’s founding HPC engineer, from 2018 through its $81M acquisition by Tempus. I built the training infrastructure behind Virchow (Nature Medicine, 2024), PRISM, and the company’s FDA-cleared diagnostic products, scaling it from 5 to 40+ ML researchers. Before that, HPC at Memorial Sloan Kettering and Caltech.
Tempus AI (formerly Paige.AI), New York City 2018–Present
Memorial Sloan Kettering Cancer Center, New York City 2018–2022 (concurrent with Paige.AI)
California Institute of Technology High Energy Physics – CMS, Pasadena, California 2015–2018
Duke University, Durham, NC — 2011–2015 — Principal Unix/Linux and network analyst on the IT Operations Management team. Owned monitoring, alerting, and change management across all academic and enterprise infrastructure; principal administrator for application performance and 24/7 NOC alerting systems.
FedEx, Memphis, TN — 2005–2011 — Senior Technical Analyst and Technical Analyst supporting Unix/Linux production systems for global logistics, SCADA infrastructure for sorting facilities, and the www.fedex.com distributed infrastructure. Started as Intern II in 2005 with the Systems Administration and Consulting group.
Contractor and university work, 2000–2005 — Building network and wifi upgrades, departmental websites, and a business backup solution for critical data.
NVIDIA DGX and GPU cluster architecture across Pascal, Volta, Ampere, Hopper, and Blackwell. Slurm, Azure CycleCloud, HTCondor, Kueue, and Kubernetes for scheduling and orchestration. InfiniBand, RoCE, NVLink/NVSwitch, GPUDirect RDMA, NCCL, and DOCA for high-performance fabrics. Lustre, Weka, GPFS, NetApp, Pure, and S3/Blob for high performance storage and data archival at petabyte scale. Azure, AWS, and GCP with Terraform, Packer, and Ansible. Network performance and systems debugging on Linux, and automation with mostly shell and Python.
Papers where I am credited for infrastructure contributions.
E. Vorontsov, et al. · arXiv preprint, 2023
A 632M-parameter self-supervised vision transformer for computational pathology, trained on a million-slide whole-slide image archive from Memorial Sloan Kettering Cancer Center. Later published in Nature Medicine (2024).
Contribution: Built and operated the GPU compute infrastructure and high-throughput storage environment used to train and validate the model.
E. Zimmermann, S. Liu, et al. · arXiv preprint, 2024
Three vision transformer foundation models trained on 3.1 million histopathology whole-slide images: Virchow2 (632M), Virchow2G (1.9B), and Virchow2G Mini (22M distilled). State-of-the-art on 12 tile-level tasks.
Contribution: Designed and operated the GPU compute infrastructure and high-throughput storage environment used to train and validate all three models.
S. Liu, et al. · arXiv preprint, 2024
A multi-modal generative foundation model operating at the slide level for computational pathology, jointly modeling histology and clinical text.
Contribution: Built and maintained the HPC infrastructure supporting large-scale whole-slide image preprocessing, model training, and validation.
E. Vorontsov, G. Shaikovski, A. Casson, et al. · arXiv preprint, 2025
A multi-modal slide-level pathology foundation model trained on 700,000 diagnostic specimen-report pairs and 14M question-answer pairs, the largest vision-and-language histopathology dataset to date. Aligns histomorphologic features with diagnostic language via clinical-dialogue supervision.
Contribution: Built and maintained the HPC infrastructure supporting large-scale whole-slide image preprocessing, model training, and validation.
Y. K. Wang, L. Tydlitatova, J. D. Kunz, et al. · arXiv preprint, 2024
OmniScreen, a high-throughput system predicting a broad range of clinically relevant molecular biomarkers across cancers from H&E whole-slide images, built on Virchow2 embeddings from 60,529 patients with paired MSK-IMPACT panels.
Contribution: Built and maintained the HPC infrastructure supporting large-scale whole-slide image preprocessing, embedding generation, model training, and validation.
H. Newman, et al. · INDIS ‘15 (SC15), Austin, Texas
Second Workshop on Innovating the Network for Data-Intensive Science, co-located with SC15: The International Conference for High Performance Computing, Networking, Storage and Analysis.
Contribution: Built and operated the SDN testbed and high-speed data-transfer infrastructure used in the demonstrations.
J. Balcas, T. W. Hendricks, D. Kcira, A. Mughal, H. Newman, M. Spiropulu, J. R. Vlimant · Journal of Physics: Conference Series, Vol. 898 (CHEP 2016)
A software-defined next-generation integrated architecture supporting high-energy physics and data-intensive science. Presented at the 22nd International Conference on Computing in High Energy and Nuclear Physics (CHEP 2016), San Francisco.
Contribution: Built and operated the SDN testbed and high-speed data-transfer infrastructure used in the demonstrations.
J. Balcas, B. P. Bockelman, D. Kcira, H. Newman, J. Vlimant, T. W. Hendricks · Journal of Physics: Conference Series, Vol. 898 (CHEP 2016)
Evaluation of HTTP as a data access protocol for the CMS Any-data Anytime Anywhere (AAA) project, comparing performance and operational characteristics against XrootD’s native protocol.
Contribution: Built and operated the SDN testbed and high-speed data-transfer infrastructure used in the demonstrations.