Recycling centers in North America reported 390 fire incidents in 2022, half of which were caused by lithium-ion batteries. My interdisciplinary senior design team and I researched this problem by talking to professionals in different parts of the recycling pipeline. The main reason why lithium-ion batteries are the cause of these fires is because of improper recycling. The batteries must have their terminals taped to stop them from igniting by short circuiting. The team has interviewed waste management professionals, battery recyclers, firefighters including Joe Dunlop, a Waste Reduction Administrator for the Athens-Clarke County Solid Waste Department. The key takeaway from Joe and other interviews is that many batteries do not come into recycling centers properly packaged, and a big pain point for the workers inside the recycling centers is to chemically sort batteries and tape their contacts, especially coin cells, due to their high volume and small geometry. 

To solve this problem, we built the BatteryBox, a novel battery recycling device which intakes cylindrical and coin cell batteries and uses computer vision to sort them based on their volatility. Lithium batteries (highly volatile) are processed differently than alkaline batteries (non-volatile) and are insulated in a layer of wax. I built a processing pipeline for image classification was using a multilayer sequential TensorFlow CNN. This classifer operates at 99% accuracy and runs fully off RaspberryPi which connects to a camera that captures images of each inputted battery. 

We used 3D printers and laser cutters to design the box out of plastic and steel alloy. The device also has a seamless user interface, for users to purchase batteries after they have recycled their old ones. 

Github | Demo Video

For the past two years, I have been closely working with a PhD student mentor at Emory University to answer the question "What are the underlying neural recruitment dynamics of motor neuron populations across varying outputs of force?" 

Throughout this process, I have worked to optimize neural spike sorting data pipelines (Kilosort, PixelProcessingPipeline) and kinematics-tracking neural networks (DeepLabCut, Anipose). The spike sorting pipelines, written in Python and MATLAB filter, sort,  and amplify spikes from raw data through gaussian filtering and dimensionality reduction to create clusters of similar spikes, identifying unique motor units throughout an EMG recording. This process requires processing gigabytes worth of data, so we use CUDA to run parallel processes on high performance GPUs. The kinematics-tracking pipeline also utilizes Python to label rodent body-parts from a 3D triangulated space compiled using four FLIR cameras. The rodents themselves are trained by us to run at varying inclines and speeds over the course of a few months to be ready to run with a surgically implanted with electrode threads in their deltoid and tricep muscles.

The outputs of these sorted motor units and tracking data are used to plot 2D and 3D representations of the neural dynamics recorded during each session. This post-processing and data handling set of functions we created is called rat-loco and is written in Python and utilizes the Plotly library. 

I began the process of writing my own thesis regarding data visualization of multi-dimensional neural dynamics when I was awarded the President's Undergraduate Research Award.

Github

During the summer of 2023, my PhD student mentor and I embarked on a project that would allow drone users to remotely fly their drones and capture footage by logging onto a webpage, creating a handshake to their drone, and maintaining a constant data connection to the drone over the internet using WebRTC, establishing a cutting-edge latency of only 100ms. We call this project, CloudChop.

The drone's flight controller was custom built using C++ and a Raspberry Pi interface, the web development thus far has been done using HTML, CSS, and JavaScript, and the internet protocol servers were set up using CoTurn. The drone prototype to the left was also prototyped by sourcing parts from the internet, the most important being the 4in1 electronic speed controller, which sends synchronized pulses to the brushless motors for stabilization and flight control.

This project was part of the Georgia Tech Create-X I2P fellowship, where we were awarded seed funding for our prototyping process, a guaranteed acceptance to the startup accelerator program, and the "Overall Best Project" award which was voted on by the other fellows in the program.

CloudChop | Github Repo - Flight Controller

As part of Georgia Tech's biomedical engineering junior design course, I led a small team in prototyping a device that could monitor the severity of wrist and finger tremors and also cluster groups of tremors by their similarity of 3D kinematics for researchers studying motor diseases.

The concept model was created using Fusion360 CAD software and the prototype was developed using 3D printed materials, an Arduino Nano, a Bluetooth chip, and accelerometers. The Arduino was pre-loaded with C++ code to process and send raw data values to a python application on the user's laptop where they can continuously monitor the severity of the patient's tremor and post-process the data to create visual representations of the tremor activity. The post-processing involved a Q-learning model to identify unique tremors based on their six axes of motion per sensor. 

I reached out to Emory's leading experts in neurosurgery and Deep Brain Stimulation research to help us design a device that would fit researchers' needs and would also be ergonomic for patients throughout a recording session. 

Final Design Video

For BMED 3600, a small team and I were tasked with creating a novel gene therapy that could help mitigate the effects of any disease. We chose the rare Type 1 Classic-Like Ehlers Danlos Syndrome, which to this day has no cure. We proposed to create an AAV vector which will contain a healthy form of the TNXB gene, which is damaged in the aforementioned disease. This vector would be injected in the afflicted areas of the body to splice and replace damaged genes.

Furthermore, we proposed to monitor the progress of patients by utilizing Western Blots and Electron Microscopy. 

Poster | Paper