- AI Supercomputers

- Team lead for Data Analytics team responsible for machine learning, numerical modeling, performance optimization, and research projects for internal and external clients.- Research, develop, and deploy machine learning based solutions for external clients including: - Developed CNN and transformer based model (ViT) to detect and classify sounds of different fish species from passive acoustic data (PyTorch, librosa). - Developed time-series based GAM to predict presence of sea ice trained on dynamical sea ice forecast model (statsmodels, Prophet, MLFlow, DVC). - Develop a CNN based model to detect coral spawning in aquaria (Fast.ai, BentoML, MLFLow, DVC). - Develop CNN based classification of images from in-situ instruments for application in a streaming pipeline (Tensorflow, Keras, Faust, PostGREST, Kafka, Avro, skimage, scipy).- Research, develop, implement, and report to stakeholders results from risk assessment studies of oil spill threats to coastal areas including: - A study assessing the risk of drifting vessels breaching and spilling oil to the Alaskan coastline. (AIS, rasterio, OpenDrift, Dask, geopandas, scipy, cartopy). - A study assessing risk of oil spills to subsistence harvesting regions along the North Slope of Alaska (Python, PyGnome, AIS, geopandas, rasterio, Numpy, Dask, gdal, shapely).- Aid in development of on-demand hurricane storm surge model deployed on cloud and HPC infrastructure (Terraform, Ansible, Prefect, SCHISM, Python, AWS).- Run quasi-operational Arctic sea ice prediction model HIOMAS including data ingestion, model pre-processing, simulation, and post-processing (Fireworks, Docker, xarray, numpy, scipy, netCDF4, POP, MPI, Fortran).- Containerize and implement distributed caching in ncWMS (Java, Ehcache, Terracotta).

- Developed and optimized high-resolution parallel 3D numerical hydrodynamic and sediment model (SELFE).- Deployed continuous integration testing for SELFE using BitBucket, Docker, and Shippable.- Developed predictive models using machine learning (scikit-learn) and statistical methods (statsmodels).

- Improved strong scaling of coastal circulation model SELFE from 256 to over 1,024 cores on leadership class HPC clusters.- Optimized SELFE using improved linear algebra algorithms, memory locality, and vectorization - 2x to 6x faster.

– Taught introduction to programming in C++ for engineering and computer science studentswithin the context of robotics.

– Developed driver for Sony Aibo robot in object-oriented C++.– Developed object recognition routines for groups of Sony Aibo robots using OpenCV.