Healthcare robotics / medical AI / real-time perception

An AI-driven ultrasound robotics project for automated prenatal screening. The system used real-time fetal standard-plane detection, segmentation, and biometry estimation to reduce the amount of manual scanning needed during an antenatal ultrasound workflow.

• Role: Led a 3-member AI/computer-vision effort as the principal model developer, coordinating with an interdisciplinary team across data, model training, integration, and deployment.
• Stack: Python, C++, PyTorch, deep learning, Segment Anything, real-time inference, robotic ultrasound prototype.
• Result: Adapted SAM into a self-prompting fetal ultrasound model for real-time use and deployed it in a robotic prototype, substantially reducing manual scanning time during demonstration workflows.
• Evidence: Showcased at the XVIII Clinical Ultrasonography in Practice (CUSP) conference, where clinicians showed strong interest in its potential for large-scale prenatal screening.

Digital pathology / deep learning / graph analysis

Deep learning research on breast cancer whole-slide images, focused on automated image interpretation and analysis of tumor microenvironment patterns.

• Role: Developed convolutional and graph neural network workflows for WSI analysis, coordinated collaboration between research labs, and mentored students working on whole-slide image analysis.
• Stack: Python, PyTorch, convolutional neural networks, graph neural networks, digital pathology, whole-slide image pipelines.
• Result: Built a multi-stage patch-level annotation model for whole-slide image analysis, supporting automated interpretation across a broad set of tissue classes. Investigated associations between neighborhood deprivation, tumor microenvironment, and race.
• Evidence: Presented preliminary findings at ATTIS 2023 and other research symposium settings.

Medical visualization / VR / real-time rendering

A real-time VR visualization system for stereo-endoscope video streams, built to make medical imaging more immersive and interactive.

• Role: Designed and developed the application from start to completion, covering video capture, image processing, post-processing, and 3D rendering.
• Stack: C++, Microsoft Media Foundation API, OpenCV, CUDA, real-time video processing, VR rendering.
• Result: Delivered a functional end-to-end system before transitioning from the company, enabling the team to later showcase it at MEDICA 2024 after competitive selection.
• Evidence: Demonstrated as a medical imaging and real-time visualization system at MEDICA 2024.

Image processing / industrial CT / nondestructive testing

An image-processing project for improving contrast in industrial CT scans, making subtle internal structures easier to inspect and analyze.

• Role: Implemented multi-scale contrast enhancement methods for industrial CT imagery and worked with application constraints from nondestructive testing.
• Stack: Python, image processing, MUSICA, industrial CT, visualization.
• Result: Improved visibility of internal structures and sharpened subtle material variations for inspection workflows.
• Evidence: Collaborated with Visiconsult, Germany, to align the enhancements with industrial nondestructive testing needs.

Robotics simulation / Unity / ROS

A 3D underwater simulation environment for testing Autonomous Underwater Vehicle software before real-world deployment.

• Role: Led simulator development for the Tiburon AUV team, building the simulation environment and software interfaces needed for rapid robotics prototyping.
• Stack: Unity, ROS, Blender, robotics simulation, sensor simulation, control testing.
• Result: Built a virtual underwater environment with simulated cameras, vehicle dynamics, sensor data, and ROS communication for testing navigation and control algorithms.
• Link: AUV-Simulator-Unity