Portfolio


Colleagues and I worked on a large, commercial inventory management prototype (designed entirely by us, built from the ground-up, and 3-D printed using an industrial 3-D printer) whose design won first place at an SXSW competition in 2017.

Photo is from the re:3D SXSW gathering.

Curated repositories list

Polycephaly

Easily create system daemons (and programs) that use an email-like syntax for communicating between threaded and forked processes.


Shoe horn

Shoe horn is a utility that simplifies the process for side-loading an OS into a VM at a VPS (which typically use KVM or XEN) via their recovery boot (which is typically a Debian-based OS, such as Finnix).

Using Shoe horn to side-load an OS onto Digital Ocean.

PyClone

PyClone is a Python package that wraps rclone and provides a threaded interface for an installation at the host or container level.


Docker


iOS

  • iOS Alerts – Properly formatted Ring, Text, and Voicemail alerts that work right out of the box.

Sneak peek

In the spirit of “release early, release often“, some of the projects that I have in the pipeline are below:

Teenage Mutant Ninja Terminals

Multifaceted project that involves hardware modification, 3-D printing, virtualization, emulation, and networking tricks. The end goal is to run software with high-demands on lower-end hardware such as a Raspberry Pi, which interfaces to 4-player controls.


Tiberium

Tiberium is a DevOps tool that I’ve been working on, which consists of three layers:

  1. Client
  2. SSH Shell (with sub-module support, such as SFTP)
  3. systemd service built with Polycephaly

Shortened training with accurate results

Some of my more recent work has revolved around Computer Vision applied to Robotics. In the interest of reducing the amount of work required to yield high accuracy for specific objects with minimal training, I devised a methodology of augmenting GPU-based training for Single Shot MultiBox Detector (SSD) in Keras. While I’m still fine-tuning the training and frame manipulation (e.g. easily adding Picture-in-picture, diagnostic info., alpha channel, and rotation) code before releasing it publicly, here are some examples:

Raw video input.

Segmentation used to isolate an object and save it with an alpha channel.

Creating training data with an alpha channel from minimal input.

The end result of heavily augmented training data yielding high accuracy.