Working alongside an internal penetration-testing team to remove business-critical privacy vulnerabilities from Ads Data Hub.Invented method for fully privatizing machine learning models using differential privacy analysis while maintaining their statistical usability.Worked alongside a research team to improve accuracy of differentially-private hyperparameter optimization for customers.Designed and developed a customer-facing feature to allow for multiple outputs from a single SQL query while maintaining privacy. Wrote monitoring and alert code to safely test and release feature publicly.

Designed video segmentation algorithm in Keras with temporal consistency capable of running at speeds of 20 FPS on Jetson TX2.Designed custom Darknet RNN model to detect objects at 60 FPS in noisy conditions on Jetson Xavier.Achieved Top 8 at the 2019 DJI International Robomaster Competition in Shenzhen, China.Trained decision tree Haar cascades with scale/rotational invariance for simpler object detection cases.

Experimented with and researched numerous data structures for privacy-safe set intersections leading to a patent application for novel approach to computing advertisers' local intersections.Contributed to open-source set cardinality estimation evaluation framework. Parallelized simulation code for a speedup proportional to number of CPU cores. Independently found and patched inefficient memory code caused by NumPy library with minimal change to codebase leading to a 60x speedup in simulation speed.

Researched and implemented SotA realtime detection algorithm trained on in-house dataset producing 20 percentage points higher accuracy (@ 50% IoU) than previously used model. Performed data augmentation steps and cleaning.Implemented drone controller and realtime 3D path planner and using OMPL and MAVROS.Worked with Gazebo simulation to quickly test and iterate on path planning design.

Created daily survey app in an agile development environment. Designed desktop UI frontend using Windows Presentation Foundation. Developed backend service to access a SQL database through Microsoft AzureDesigned and trained a machine learning algorithm in Keras to upscale images by 4x while increasing image quality during OneWeek Hackathon.

Developed Virtual Reality game for the HTC Vive using Unity3D and Blender. Worked on versatile state machine AIs for predictable enemy behaviors. Designed and produced realtime building destruction physics under strict 90 FPS performance constraints.