Battery Testing Simulator with Real-Time CAN Encryption
Using an STM32, I engineered a device that can be used to simulate a charger or a battery to evaluate performance and functionality. The simulator ensures the integrity of the data through real-time encryption of CAN communication and can simulate commands using the UART interface.
Low cost dirt detection system using spectral material analysis
Developed as a research project for Kärcher’s new fleet of products, I engineered a machine learning algorithm that used data from a low-cost spectrophotometer to classify dirt particles in real time. The algorithm could then be used to intelligently switch between cleaning modes and improve efficiency.
Industrial-grade wireless sensing network
Using industrial grade Bosch sensors present on the Nicla Sense ME, I developed a communication network that collected temperature, humidity, and movement data. In addition, I focused on reducing the device’s footprint and energy consumption.
This cross-platform web app helps the visually impaired in navigating and finding objects in their surroundings. Created as part of the MIT FutureMakers Create-a-thon, I optimized the object detection, speech recognition, and navigation algorithms to run on edge devices without internet.
Fall detection - ML on the edge
Leveraging the power of machine learning on edge devices, I was able to provide affordable and reliable fall detection for the elderly. This project consisted of engineering a neural network using deep learning to detect falls in real time using a 9-axis accelerometer and optimizing it to run efficiently on a microcontroller with 95.81% accuracy.
This app encouraged eco-conscious behaviors by reducing the reliance on single-use plastic water bottles, winning first place at Jugend Forscht. This led to a positive impact on the environment as well as significant cost savings for users. The app aided outdoor-enthusiasts find water fountains in their surroundings, helped them navigate to the location, and tracked fountain use over time.
Developed before the current wave of generative AI algorithms gained widespread attention, I developed a Convolutional Neural Network (CNN) trained on an extensive dataset of over a million images to create art using neural style transfer. This initiative was used as an incentive to collect over $500 in less than two months for the Black Lives Matter movement.
Built using the blazing fast open-source static site generator Hugo, my website is a showcase of personal and professional projects, as well as a blog where I share my thoughts and insights on various topics. This website pushes the boundaries of performance, elegance, and design, achieving a perfect score across the board on Google’s PageSpeed Insights.