Hi!
I am skilled at the following languages/frameworks
I am proficient at the following languages/frameworks
I am familiar with the following languages/frameworks
I am currently a Software Engineer in the Azure AI Speech team, part of the Microsoft AI Platform organization.
Currently, I am leading the following major efforts:
1. Increasing the training speed, performance, and size of deep speech recognition/language models by applying state-of-the-art sharding, networking, and architecture-based techniques
2. Maintaining and refining the framework that the Azure AI Speech team uses to train models
3. Optimizing model inference and quantization (for ONNX and PyTorch)
4. Using training metrics and PyTorch/NVIDIA/ONNX profiling tools to understand inefficiencies and bottlenecks
5. Maintain and develop a distributed deep learning framework for training large, deep models
I returned to the robotics team at Unity to integrate inverse kinematics directly into Unity for 5, 6, and 7-DoF robotic arms. I also implemented joint controllers to model realistic robotic behavior and engineered a VR experience to capture a robot’s workspace in Unity. I found myself using a variety of mathetmics skills, including linear algebra, calculus, trigonometry, and pre-calculus methods.
I worked as part of the gamer serivces team in the GeForceNOW (streaming gaming) organziation. As an intern, I proposed a complete design and architecture that enabled GeForceNOW searching to be operated completely in the cloud, as powered by AWS Elasticsearch. I developed the infrastructure, design documents, documentation, and searching algorithms needed to power searching across a variety of clients. I became highly experienced in AWS and Elasticsearch as well as GraphQL and Python.
I worked in Unity's AI department as part of the new robotics team in order to integrate motion planning and inverse kinematics for robotic arms (e.g. UR3) into Unity. This included working with GoogleX's robotics team quite closely. I also helped to link ROS and the Unity simulation engine via a TCP endpoint. In addition, I explored and implemented both classical and machine-learning driven robotic manipulation techniques in Unity, including pick-and-place. Lastly, I integrated my work with a pose estimation neural network (PoseCNN) in order to pick-and-place unknown objects.
As part of TAMID Group's annual summer fellowship program, I was given the opportunity to intern with Aidoc Medical in Tel Aviv, Israel. During my 8-week internship, I was responsible for streamlining and automating a variety of processes involving the medical data Aidoc receives from its partner hospitals. Much of my work centered around updating existing Aidoc infrastructure to utilize Amazon Web Services, including EC2, EFS, EBS, and VPC. Specifically, I proposed a scalable yet economically-viable cloud processing and storage solution that I implemented with my fellow interns. Upon the completion of our work with AWS, I worked with Jenkins, an automation server, in order deploy these updated processes in an accessible manner to team members within the company. This included catering to the needs of users by creating a practical user interface in Jenkins. Most of the AWS work was done in either bash script or Python (and its many libraries, including Tensorflow and Paramiko), and Jenkins is a Groovy-based interface.
During my 3-month internship, I helped Strayos implement a role-based action control (RBAC) protocol
within their web application using full-stack knowledge, which included applying Ruby, Rails,
Angular,
RxJS, and HTML skills. As a part of their collaborative team, I was tasked with working with a
variety
of different team members, who each possessed their own set of skills as well working on customer
and
internal deadlines. I was also tasked with creating both internal and end-user documentation, as
well as
using SQL, PostgreSQL and SSH protocol to migrate existing user data on our platform.
Strayos can be found here
and my documentation can be found here
Through the University of Missouri-St. Louis' STARS program for talented rising high-school juniors, I was matched with Saint Louis University's Parks College of Engineering, where I worked with Dr. Mark McQuilling concerning a new polysonic wind tunnel. Using MatLab, I performed data analysis on information collected from the wind tunnel, and I became experienced with specialized procedures such as data entry and wavelet transform. I concluded the program but presenting my work at a small conference in St. Louis.
For my undergraduate thesis, I built a DC-GAN capable of creating novel hand X-Ray images from the RSNA dataset. I then attempted to improve the accuracy of an existing bone-fracture classifier. My paper can be found below in Other Work.
I am the co-founder of the Wharton Undergraduate Aerospace Club (along with Ethan Markwalter), which serves to educate and promote the aerospace and white-space industry at Wharton in creative, hands-on ways. Our current website can be found here.
I am currently developing a robust neural network to detect maintenance anomalies in commercial aircraft. Anomalies include metal fatigue and fuse-pin misalignment, and adversarial defense techniques, including PGD, are used. My current code can be found here here.
Current
I lead the 100-person Penn Aerospace
Club. This includes overseeing four sub-teams and coordinating launches.
Past
During my freshman year at Penn, I worked
as a part of Penn's high-altitude balloon team where
I worked with an Iridium Module and an Arduino Board to establish communication between team members
on the ground
and the airborne balloon. My progamming was primarily done in both Arduino and the
Iridium API, and we had three successful launches using that code, which can be found here.
Following my work as the Satellite Team Lead, I was selected to lead the entire high-altitude balloon team along with two other students. Because of the logistics of our lauches, this leadership position requires deft coordination between all of our members, as well as the entirety of the aerosapce club. Under our leadership, we launched 6 times during the 2018-2019 school year, including 2 successful launches that reached 60,000+ feet.
I am working on an literature review on DNA-based neural networks. This includes recent work on implementing both Hopfield and winner-take-all based systems in DNA. My paper can be found here.
I am currently working on an iPhone application that can monitor a user's form for a given exercise (e.g. squat) and give feedback. This application has the goal of helping new lifters and lifters who prefer to lift alone, as well as preventing injuries. My current code can be found here.
As part of my operating systems course, I developed an operating system from scratch in C, along with with 3 other students. I implemented the scheduler from scratch using the ucontext library, and I also created most of the kernel logic. Code is avaiable on request as it is not allowed to be public.
I created a complete pick-and-place pipeline for my robotics course, along with 2 other students. I designed the actual picking and placing using inverse kinematics and an approach/retreat calculation. I also helped in implementing the RRT* algorith used to move from the blocks to be gripped and the platform they were to be placed on. This project was done in Python using Gazebo and ROS.
I am undertaking advanced economic time series forecasting using R as part of Penn Economics in order to better understand the US automotive market. My current code can be found here.
I have implemented a variety of advanced computer vision algorithms in a simple Python library. My current work can be found here.
I created a Google Chrome extension that allows users to highlight text and put it into a new or exisitng Googl Sheet for easy note-taking. My code can be found here.
During my sophomore year at Penn, I designed a Python library that implements the RankSVM machine learning algorithm to predict the final standings of an NBA regular season using only statistics gathered from previous years. Specifically, I designed and implemented the data pipeline to transform the CSV data into data that the RankSVM could utilize. A brief presentation on our work can be found here and my code can be found here.
As a part of my cloud computing class, I worked on the front-end of an app which implemented the basic functionalities of Facebook. Besides designing the entire user interface, I also aided in designing the Map Reduce job used to suggest friends to users based on their mutual network of friends. My code can found here.
The final project for my computer systems class was to design a compiler for the J language in C. This design required a variety of special-purpose structures, proper memory managment, and the ability to design code in modules for easy integration. My code can be found here.
Degrees: Bachelor's of Science in Engineering
Master's of Science in
Engineering (in robotics)
Majors: Computer Science and Economics
Relevant Coursework: Intro to Machine Learning (Graduate Course),
Networked Systems (Graduate Course),
Database Systems (Graduate Course),, Computer Vision (Graduate Course),,
Cloud Computing, Econometrics, Game Theory, Mathematical Statistics,
Principles of Software Engineering,
Extracirriculars: Penn Aerospace Club, TAMID Groud, Phi Kappa Psi, Penn for
Liberty
GPA: 3.93/4.00
Awards: Excellence in Research (STARS 2016; for wind tunnel research performed
with Dr. Mark McQuilling); All-State (Track and Field 2015); MICDS High Academic Honors
(all four years);
University of Rochester's Frederick Douglass and Susan B. Anthony Award (2015; for my
commitment to social justice and high academic grades in the hummanities and social
sciences)
Extracirriculars: Progamming Club, StuTech, Community Service Committee, Track
and Field (4-year varsity),
Football (3-year varsity), Asian Cultures Club, Robotics, Election Club (2016), TEAMS,
Aviation Club, Investment Club, International Council, Cum Laude Society
GPA: 3.75
Information: During the first semester of my junior year at Penn, I did an
exchange at ETH Zurich in Switzerland, taking all graduate courses.
Relevant Coursework: Reliable and Interpretable Artifical Intelligence
(Graduate Course), Monetary Policy (Graduate Course), Financial Market
Risks (Graduate Course),
Wireless Communication Protocols (Graduate Course), Computer Architecture
(Graduate Course),