- Worked on compiler used by TensorRT for optimizing neural networks on GPU

- Implemented graph rewrites for Haskell compiler targeting in-house deep learning accelerator- Developed an embedded domain specific language for fast prototyping of hardware instructions

- Developed tensorflow graph transformation tools as a frontend to the compiler for in-house AI accelerator- Implemented custom ops and quantization kernels for accuracy benchmarking with different precision- Worked on circuit physical design tool using SMT solvers in Haskell

- Implemented a traffic sign recognition convolutional neural network using tensorflow in Python and ported to embedded C++ framework to run on dedicated vision processor- Integrated a C# driving simulator with vision processor to evaluate performance of neural networks- Improved speed of application by enabling multi-threading and maximizing parallelism- Experimented with different LiDAR neural network publications using simulated 3D point clouds data

- Developed object detection and breathing monitoring applications for ultra-wideband sensors using Python for interfacing on drones; presented results using dynamic heatmaps and graphs using matplotlib- Improved breathing tracking accuracy using Fourier transform and filters from Scipy library- Implemented secondary object detection capabilities using OpenCV Aruco markers for validation

Worked with the augmented reality headset team.- Expanded classifier framework for predicting object detection and tracking performance by implementing k-fold cross validation and feature engineering- Developed Python scripts to graphically illustrate performance of classifier, such as ROC curves

Worked on creating unit tests for web applications using UI automation tools