Student Cluster Competition

The Student Cluster Competition is a yearly competition where teams from universities across the globe compete to build the fastest supercomputer that uses less than 3,000 watts. We have competed three years and have always had very different systems to the norm. While most teams bring an Intel/NVIDIA solution, we try to use AMD systems. The second year we competed our system was lost in the mail so we built a supercomputer from left over show-floor systems and came 9th of 15 teams. The experience was one of the most stressful but educational moments I've had to date.

OpenCL over IP (CLIP)

While working at the Northeastern University Computer Architecture Research lab (NUCAR), I had an idea for a distributed GPU compute solution that used commodity networking hardware. Previously, GPU clusters needed Infiniband or other expensive hardware to work correctly and they required an immense engineering effort as technologies such as MPI needed to be used. CLIP is a simple C library that made GPU-powered applications instantly able to use GPUs on multiple systems with very little changes to the code. It also allowed systems without GPUs to use GPUs on shared systems to speed up workloads. Read more about it in the published paper linked below.

Creative Design

Hetero-Mark is a benchmarking suite that exploits many of the features of the Heterogenous System Architecture (HSA) 1.0 specification. The suite includes applications from a number of problem domains including signal processing, cybersecurity, machine learning, and data analytics. The suite is continually under development, and will include source code in multiple languages. The current suite includes benchmark source code in OpenCL 2.0 and HSA HC. I personally wrote the cryptography test suite from scratch, which runs parallel versions of the AES-256 benchmark.

UBER Pointerizer

UBER Pointerizer was a personal project where I reverse-engineered the UBER API to track the movement of vehicles in a city. Based on where cars would "disappear" and "reappear" I was able to infer trips that had taken place in a city.

  • Date: December, 2014
  • Fields: Reverse Engineering, Programming
  • Link to Project: UBERPointerizer

Project Holeshot

Project Holeshot was a self-driving RC racing car that was the winner of the Microsoft and NVIDIA prices at HackUMass 2015. We used an NVIDIA Jetson and an Arduino to control an off the shelf RC car then controlled it using OpenCV and computer vision techniques.

  • Date: October, 2015
  • Fields: Hackathon, Hardware, CUDA, OpenCV

SafeStash

SafeStash is a facial-recognition medicine cabinet created at HackBeanpot 2017 that won "Most Technically Advanced Project". The goal was to reduce prescription medicine abuse by locking separate compartments in a cabinet and unlocking them if the face matches the owner of the medicine. We used the Caffe deep-learning framework to perform the facial recognition and created a Spring Java server that controlled the cabinet from a central server which allowed the cabinet to be powered by a inexpensive Raspberry Pi Zero.

  • Date: January, 2017
  • Fields: Hackathon, Hardware, Software, Deep-learning, Facial Recognition
  • Hackathon: HackBeanpot
  • Project Repository: SafeStash

Spire

Spire is a tool I am currently writing to ease cloud and physical server deployments. You can provide a configuration for a server and spin new instances up easily in AWS, Azure, or locally. Systems can be managed, tasks can be started, and costs can be visualized from a central source. It is still being worked on so stay tuned!

  • Date: 2017-2018
  • Fields: Software, Cloud, VM, Deployments, Cluster, HPC, Datacenter

PyASN.1 PER Support

A project I was working on at an internship manipulated messages encoded in ASN.1 PER (Packed Encoding Rules). They had a huge suite in Python to manage and generate these messages but no way to export or import them in the PER encoding. So I implemented the PER standard in Python from the official specification and then worked to allow my code to be released as open-source. Eventually I want to roll my encoder/decoder into the official PyASN.1 library.