Technical lead for building out an experimentation platform for deep learning model benchmarking & analysis
Comparing results between NVIDIA GPUs and Lightelligence Optical Processing Units (OPUs)
Variety of other tools for interdisciplinary teams
Contributed to development of initial OPU SDK, particularly in API design and refactoring
Conducted feasibility studies of different model architectures for deployment onto OPUs
Promoted to Lead Engineer and Architect of IBM’s end to end activity detection system for the DIVA project
Developing Frater, a machine learning and data-driven system framework for modularizing pipelines into components to allow for more flexibility
Published research in continual learning for object detection models to allow for incrementally learning new object classes
Trained and tested activity detection models for spatio-temporally localizing and recognizing activities in surveillance video for the IARPA DIVA Program
Collaborated with researchers from MIT and Purdue University to build a state of the art activity detection system and publish papers to top conferences
Designed and developed an end to end activity detection system for running in a deployed environment
First place in the TRECVID Activity in Extended Video 2018 competition
Developed Tensorflow and Scikit-Learn machine learning models to detect and predict container failures in Kubernetes clusters.
Designed and architected machine learning pipeline for data cleaning, feature extraction, normalization, learning, and prediction
Containerized Go and Python applications into Docker images to be deployed into a Kubernetes cluster
Worked on an Agile team and practiced continuous integration and continuous development using GitHub and TravisCI
Developed a 4-minute pitch every week with project team to present to IBM executives