Using an Android Things developer kit I received at a Hackathon, I created a home assistant which caters to my house.
The 8-segment display which shows the time until the next inbound T (Boston subway) will arrive to the nearest stop to the house.
Utilizing the camera, the device detects faces using the Cloud Vision API, and upon seeing a face, greets it by saying "Welcome home". Currently, I'm working on integrating facial recognition, such that the device will be able to say "Welcome home [name]" for stored faces.
Current work also involves integrating the Google Assistant API.
Working under Professor Holt, a student and myself were tasked with improving a sonic levitation demonstration. 4 speakers serve to levitate a styrofoam ball in a chamber. When running, the apparatus is very loud, and requires a large setup with bulky frequency generators and amplifiers.
My part of the project involved simplifying the large frequency driver setup to a smaller, portable system. To accomplish this, I moved frequency generation to small ATTiny85's paired with Chebyshev filters (based upon this). Then, a parent Arduino controlled 4 of these generators to ensure their phases were linked.
For a final project in an introductory Python course, I worked together with one of my peers to create an implementation of a Neural Network from scratch in Python, using Numpy. We used the common MNIST handwritten digit database, and set out to create a network to recognize digits.
Our network used the Stochastic Gradient Descent, along with optimizations like minibatch for training. We also implemented a basic form of annealed learning. Our network reached a peak accuracy of about 96%.
Group project in learning methods for machining different materials. Made from acrylic and HDPE. Uses spring and moment arm mechanism for lever action.
Roasted single-origin beans on a stovetop. Kept log of roast time, temperature, and color. Blind tasted each batch and compared to data, qualitatively verifying the relationship between roast time and certain flavor characteristics.
3D printed mold, mimics elements of a metal casting mold, uses wax. More successful design created allowed part to be removed (not pictured). To be used as a teaching tool.
Given challenge to create a simple stand from a steel bar, allowing for two viewing angles of the clock. Design only uses bends, no welds or joints.
More not pictured: