With a background in Biomedical Engineering, I am deeply committed to advancing research in computational neuroscience, brain-computer interfaces, and neuroAI. My passion lies in exploring how neuroscience principles can be applied to build smarter, more adaptive technologies. I aim to contribute to research that deepens our understanding of brain function and cognition, and translates these insights into real-world applications, such as neuroprosthetics, assistive technologies, and intelligent systems. My focus is on integrating neurodata and leveraging the intersection of AI with neuroscience to drive innovations that bridge the gap between human cognition and machine intelligence, ultimately enhancing quality of life and expanding human capabilities.
My interest in Brain-Computer Interfaces is rooted in my desire to develop direct communication pathways between the brain and technology. Through my work with neurodata and custom machine learning models, I aim to create solutions that improve assistive technologies and neuroprosthetics, especially for individuals with motor impairments. I am fascinated by the potential of BCIs to enhance human-machine interaction and empower those with limited mobility.
Biomimicry inspires me to look at natural systems for innovative solutions to engineering challenges. My background in designing medical devices for low-income settings has shown me the importance of efficiency and adaptability—qualities that nature exemplifies. I am particularly interested in how biological strategies can influence the design of sustainable medical technologies and improve device performance, especially in resource-constrained environments.
My passion for NeuroAI stems from my work in computational neuroscience, where I have explored brain connectivity and neural data analysis. I am driven by the idea of using brain-inspired models to advance AI capabilities and create intelligent systems that learn and adapt like the human brain. By integrating neural data with AI, I hope to contribute to innovations that push the boundaries of neuroprosthetics, intelligent interfaces, and cognitive technologies.
This project investigates the computational benefits of expansion and sparsity in neural systems for processing clustered sensory inputs, inspired by Babadi and Sompolinsky's 2014 study. The focus is on how expansion (increasing the number of neurons in downstream populations compared to input pathways) and sparsity (few neurons being active at a time) affect variability and noise in clustered sensory data. Unlike the original study, which used a binary classification task, this project applies these principles to a linear regression task to further explore the computational advantages of this approach.
This project addresses premature death caused by respiratory distress syndrome and malnutrition in preterm babies, particularly in low and middle income countries (LMICs). Due to immature neurophysiological functions, preterm babies cannot coordinate their own feeding, leading to a high reliance on healthcare professionals for manual feeding, as affordable automatic systems are lacking. The project's goal was to design an automated feeding system that accurately delivers breast milk to premature infants while considering local constraints, reducing the burden on healthcare workers and improving the precision of feeding. The system allows clinicians to input flow rate and volume, which is processed by a microprocessor to deliver the nutrition accurately. The design includes an LCD, sensors, LEDs, a buzzer, a peristaltic pump, and a microprocessor. It successfully delivers breast milk at flow rates between 5-15 ml per minute (with a maximum error margin of ±3 ml) and volumes between 20-60 ml.
This project automates the detection and labeling of lanes in 96-well plate gel electrophoresis images to streamline the analysis of large datasets. This involves utilizing image preprocessing and object detection techniques with OpenCV to identify bands, generate bounding boxes, and calculate lane numbers based on ladder positions. The algorithm was tested on 3000 images, achieving an 83% success rate on images with lane widths of 9 and 10 but was less effective on images with broader lanes. Enhancing the algorithm's adaptability to various lane widths could establish it as a valuable automated tool for accurately labeling gel electrophoresis images, thus optimizing the efficiency of biological data analysis.