Part of Launchpad, an ML/AI club that focuses on creating unique solutions with different techniques. Worked on projects for: Car Path Prediction, Food Image Generation, Basketball Playoff Prediction, Low-Cost Embedded Animal Classification.

Leading the club, accountable for overall progress made over the course of the summer and semester. Providing a vision for the club's direction, including revamping website, changing recruiting process, brainstorming how to conduct club activities virtually, overseeing progress of all ML projects and fellow officers.

Led 2 projects over the course of 2 semesters, involving finding research topics, leading a team of 5-8 people, writing our results, directing and managing the vision of the project.Project 1 - BalliNN: Brainstormed, implemented, and tested various ML models to see if we could accurately predict the winners of different playoff series. Got to a best prediction of 79% with logistic regression and custom pruned data.Project 2 - Backyard IOT: Taught a team of 5 about embedded development and computer vision on IOT, while building and training a fast, efficient ML algorithm (retrained MobileNet-V2) that can accurately predict the classes of several kinds of animals on a Raspberry Pi at over 5 frames per second.

Created CNN based encoder-decoder model to refine and smartly interpolate depth estimation to fit current RGB image data. Built training and PyTorch to TFLite conversion scripts in Pythonand deploy pipeline in C++ to reach better accuracy at ~100 FPS. Integrated refinement into company SDK as a local API call.

Trained ML models to identify basketball plays given tracking data and other specialized model outputs. Analyzed player tracking data to inform strategy on player positioning while defending shots. Developed universal features in Python to track across different models that resulted in >5% accuracy boost across the board. Honed important edge case performance through active learning and class balancing to meet ROC-AUC benchmarks for clients.

Built an API network that facilitates telemetry and health across a voice platform that is scalable to many services. Created a flexible, robust and quick architecture under MVC that future teams can build upon, with logging and a basic frontend.Made with C#, ASP.NET Core, Azure.

Worked on the Jetson Nano platform to brainstorm and program smart features for the ROAR race competition for RC cars, including: mapping with QR code based localization and global position translation with rotation matrices, ground-plane detection for obstacle avoidance, and keypoint matching across frames for real-time position difference localization.

Built an end-to-end pipeline for connecting vision based AI solutions to an easy to use cloud interface on ARM64 and x86 devices for both developers and enterprise. The former is currently open-sourced. The pipeline was built using Azure IOT Edge, Docker, OpenCV, Multi-Threading, Requests/CPPRestSDK, and Azure Cloud Storage on both Python and C++.

Using the OpenPose API for pose-estimation combined with a custom neural network written in Python using the TensorFlow framework, 3 other interns and myself were able to help a robot (Pepper NAOqi interfaced with using C++ and the Libqi API) understand gestures, particularly waves, to help facilitate an actual conversation. Group also experimented with RNNs/LSTMs written in PyTorch.

I volunteer bi-weekly at the Tech Museum teaching the community about the value of STEM in our everyday world.

Oversaw the operations of the entire robotics organization, comprising of 12 robotics teams.

Managed and trained the entire saxophone section of the school band. 1st chair and principal.

I introduced computer vision to my robotics team, and since then I've implemented it using OpenCV and Python on the Raspberry Pi. Currently, we are working with deep learning inference on the edge on the Jetson TX platform and bringing it to Cupertino Robotics.

Created and mentored 2 FIRST Lego League middle school robotics teams through CHS Robotics comprising of about 10 students each. Since expanded into 3 teams.

Assisted in overseeing the entire CHS Robotics organization.

Using the Keras, Tensorflow, and Caffe frameworks on C++ and Python, I built neural networks that can teach a RC Car to steer and detect people for the purpose of aiding fire evacuation in an efficient and safe manner. The networks were constructed using a concatenation of 2D LiDAR and camera image data for steering, and image data for people recognition. The RC Car was interfaced with using a Jetson TX2 running Linux4Tegra. For more information, check out the Github repository here: https://github.com/NVIDIA-Jetson/Formula1Epoch

Mentored fellow swift programmers at a hackathon.