Projects Personal Projects
My dedication to writing software and devising solutions goes beyond my professional responsibilities
Digital display replica for an F/A-18 simulator
A project that challenged my ability to design both software and hardware to accomplish a fully functional product. I had to design a custom PCB in KiCAD that could fit around the edges of an LCD display in a slim frame. But that was only the start of it. After carefully selecting components that could fit on the PCB, I designed a 3D model from scratch using Fusion360 which could be 3D printed to house all of the internals. Finally, I wrote a C++ program for an STM32F103 microcontroller to interface with a simulator (DCS World) via USB. The final result was a huge success and a personal win, as I sold these units to simulator enthusiasts. The project took many prototypes to get right, in particular, the intricacies and difficulty of 3D printing to exact dimensions.
Speech Emotion Recognition Neural Network
A python GUI tool that utilizes Convolution Neural Networks to classify the emotion of a voice from a short clip recorded by microphone. The tool contains a pre-trained model but also allows the user to further fine tune the model on their own voice samples or a set of .wav files. The tool provides a hyper parameter tuning as well as producing a transcript of user's recordings.
WiFi Tracer Unit Chronograph
Another project intended to fulfill a niche within the airsoft community. Using my skills from the F/A-18 DDI Replica, I designed a tracer unit which is a device that produces a bright flash to illuminate incoming glow-in-the-dark projectile, causing it's trajectory to be lit up at night. The device was WiFi enabled (using an ESP82666 chip) and could detect the speed of a projectile (chronograph). The software (written in C++) creates a WiFI Access point and hosts a HTTP web server that feeds live data (using WebSockets) to the user's smartphone, providing unit statistics, battery, and of course the chronograph. I also designed it using Fusion360 (for 3D printing) and KiCAD for PCB design, this time utilizing mini SMD components.
Recurrent Neural Networks for Gyro Sensor fusion • Dissertation
Paper I wrote for my Bachelor's thesis combined my love for quadcopter racing and machine learning in order to solve the challenge of attaining accurate estimations from MEMS IMUs sensors. Getting accurate data from MEMS is a well-known challenge due to gyroscopic drift, accumulation of errors and non-linear scale factors. In the paper I discuss sensor fusion solution is using Long Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) recurrent neural networks to estimate quaternion-based AHRS orientation of a low-cost MPU9255, gyroscope accelerometer, magnetometer. The paper compares sensor fusion approaches by Madgwick and Mahony to LSTM and GRU neural networks.
