Empowering neuroscientists with cutting-edge AI, I developed a web-based neural circuit segmentation tool, as part of my Ph.D dissertation at Harvard. This interactive platform leverages real-time segmentation neural networks and object tracking to automate the analysis of complex biological structures. Researchers can upload their data and witness immediate visual feedback, with the tool simultaneously segmenting objects of interest and providing powerful visualizations of both the input and output. This streamlined approach significantly reduces labelling and analysis time and empowers researchers to delve deeper into the intricacies of neural circuits.

Technologies used: Javascript (ThreeJS), Python Flask, Neo4J, and TensorFlow

Streamline Neural Circuit Reconstruction with Our Interactive Proofreading Tool. This web-based platform, developed as part of my Ph.D dissertation at Harvard, empowers neuroscientists to accelerate the quality and efficiency of 3D neural circuit reconstructions. Utilizing real-time object tracking, the tool pinpoints segmentation errors within complex datasets. Furthermore, researchers leverage the powerful 3D visualization capabilities to swiftly inspect and rectify inconsistencies. This innovative framework fosters not only efficient proofreading but also a deeper understanding of evolving neural graphs and facilitates the sharing of error correction strategies within the research community.

Technologies used: Javascript (ThreeJS), Python Flask, Neo4J, and TensorFlow

As part of my research work at Harvard Business School, I develop simulations that utilizes cutting-edge environments to create realistic traffic scenarios. This allows us to analyze consumer behavior and attitudes towards various self-driving car functionalities in a safe, controlled environment. This innovative approach provides invaluable insights for marketing self-driving cars, highlighting their capabilities to a hesitant public. Furthermore, the project generates valuable data that can inform the development and refinement of autonomous dvehicle technology, ultimately accelerating the path towards safer and more reliable self-driving cars.

Technologies used: Unreal Engine 5, Python, Qualtrics, Javascript

The image relighting tool empowers real-time manipulation of synthetic lighting parameters (position, color, direction) to enhance photos, particularly for darker skin tones often compromised by poor lighting. This innovative approach corrects inadequate lighting and produces natural-looking results, not only overcoming photography limitations but also opening doors for photo editing software, virtual try-on experiences, and accessibility tools for low-light environments.

Technologies used: Javascript (ThreeJS), Python Flask, and TensorFlow

Juba Face ID, an Android app, developed for University of Juba, streamlines student attendance with facial recognition. It leverages a two-step process: first, MediaPipe, a powerful open-source framework, detects faces within captured images. Next, a custom-trained FaceNet model, the core of Juba Face ID, extracts a unique facial signature from the detected face. This signature is then compared against a database of registered students, marking attendance for successful matches. This innovative app simplifies attendance tracking and enhances security.

Technologies used: Android SDK, MediaPipe, Python, and TensorFlow

As part of my master's thesis, I developed an interactive toolkit for deep neural network training in neuronal structure segmentation. This tool tackles the challenge of extensive manual annotation by employing a dynamic feedback loop. Users provide sparse annotations through a graphical interface, and the system iteratively trains the network on these annotations, displaying near real-time predictions. This collaborative approach not only reduces annotation burden but also facilitates parallel annotation by multiple users, all receiving feedback from the evolving classifier. This project paves the way for more efficient and collaborative deep learning in the field of neuroscience.

Technologies used: Javascript (ThreeJS), Python, and Pytorch