UofT CubeSAT - 2014 to 2016
I was the head of the attitude determination and control systems division on a University of Toronto team building a small satellite. I designed the part of the satellite that would keep it in a stable orientation to meet our communications, power, and thermal requirements. Much of my work consisted of writing simulations in Matlab that modelled the changing orientation of the satellite as it orbited the Earth under various environmental disturbances, such as differential gravity and aerodynamic drag.
An iOS app that let's you easily define electric circuits consisting of voltage sources, current sources, and resistors then automatically solves for the electric potentials at every node.
CIRCUIT SOLVE - 2015
An iOS app that helps groups track and split shared expenses. A typical use case is when you go on a trip with a group of friends and you all start paying for drinks, food, etc. At the end, SplitIt can tell you who owes who and how much, resolving everyone's expenses in the most efficient way.
SplitIt - 2015
MAPPING APPLICATION - 2016
A Google Maps-like application built from scratch in C++. It interactively displays information such as streets, street names, points of interests, and terrain information such as lakes and parks. Like Google Maps, it does this dynamically depending on the zoom level so that information is never too cluttered. You can also search for streets or places using natural language and find the best directions to them. The map also addresses the travelling salesman problem using simulated annealing and other techniques, meaning you can enter many locations and find an efficient route that traverses all of them.
SPATIALLY AWARE AI - 2017
RAY TRACER - 2017
A ray tracer built from scratch in C++, which can generate still images with high fidelity to the real life properties of light. Features include anti-aliasing, soft shadows, glossy reflections, refraction, texture mapping, arbitrary mesh rendering, customizable material properties, skyboxes, bounding-box hierarchies, and much more.
An artificial intelligence that produces it's own original piano pieces by learning from midi files. It uses deep belief nets and recurrent neural networks to learn what makes good music, including tempo, key, chord progressions, and more without any specific guidance. So far, it's been trained on pop, jazz, and classical music. Soon, it will also be trained on pieces from my favourite band: Radiohead (once I've finished transcribing enough of their songs to midi files!).
AI MUSICIAN - 2017
A CPPN network that can generate original abstract art at any resolution. It uses a deep neural network to make complex functions that define patterns and fractals, which when executed pixel by pixel result in some pretty beautiful pieces (at least in my artistic opinion).
COMPUTER ABSTRACT ART - 2017
An artificial intelligence that can determine, to a certain precision, both its position and orientation relative to a starting point given a sequence of images. It does this only using vision and learning, meaning that there are no sensors or carefully designed algorithms involved. The AI consists of a deep convolutional neural network and the training data was generated using my very own ray tracer. As far as I am aware, this is a novel technique that has a lot of potential for the growing augmented reality industry, where spacial awareness is key to nearly all applications.
Other projects I've worked on
-
Financial reporting websites at Orbis Investments
-
A basic operating system with processes, threads, and virtual memory
-
A basic end-to-end C compiler
-
Dāna, an app to make donating to local charities and shelters easier