S K IL LS
Reinforcement Learning, Deep Learning, Time-series analysis, recommendation systems,
regression analysis, graphical models.
Engineering: AWS, GCP, Tensorflow, SQL, BigQuery, golang, python, sklearn
- Created, tested, and deployed reinforcement learning models for email personalization.
- Lead product-placement recommendation engine project, resulting in a 6% increase in click-
through-rate at 95% confidence.
- Leveraged various technologies within GCP such as deploying tensorflow models to bigquery
databases in order to execute massively parallel inference computations.
Designed, tested, and deployed algorithms for converting accelerometer data streams into real-
time activity detectors/classifiers
- Each detector had an extremely low failure rate, with 0.99 F1 scores on average.
- The success of the algorithms was crucial to earning Kinetic a 4.5M seed funding round.
- The models varied significantly in complexity, from linear regressors to recurrent neural
nets with convolutional layers. Data labeling was done via Mechanical Turk, research was done via
a python datascience stack, and production code was deployed in Golang.
- Collaborated extensively with the engineering team to create an in-house pipeline to convert
CNNs trained in tensorflow on AWS deep-learning instances into json files, enabling models to run on an ARM64 architecture.
Focused on research related to autonomous vehicles, including pedestrian forecasting.
Researched numerical methods for diffeomorphic image registration, for various downstream
applications such as improving detection of Ahlzheimer's disease.