To learn is to live.
Ph.D. Candidate in Computer Science
University of Rhode Island
GeoAI · Computer Vision · Sustainable AI
I’m Chan Young Koh, a Ph.D. candidate in Computer Science at the University of Rhode Island.
My research connects computer vision, GeoAI, and robotics to make pedestrian infrastructure more accessible. I build systems that detect objects, assess sidewalks, and bring spatial intelligence onto robots.
I also study context-aware sensor analytics and the environmental footprint of LLM inference. In the classroom, I focus on clear examples, structured practice, and helping students become independent problem solvers.
Transformers · DETR · YOLO-OBB · Segmentation
ROS · RGB-D · LiDAR · SLAM · NVIDIA Jetson
RAG · FAISS · Sensor-pattern embeddings
vLLM · llama.cpp · Ollama · SGLang
Spatially reliable object detection for GeoAI and accessible pedestrian routes.
Multimodal data fusion for sidewalk infrastructure assessment.
Interpretable anomaly detection in mobile and urban sensor streams.
Hardware-aware benchmarking of the energy, water, and carbon footprint of AI.
Data analysis with clustering, visualization, and practical evaluation.
2025–Present
Spatially Preserving Object Detection with Transformers
A compact, DETR-style detector for GeoAI and spatial sensing. I preserve weak but meaningful object features before decoder refinement to improve boundaries and separate adjacent instances, with evaluation across COCO, VisDrone, and road-sidewalk transfer settings.
2025–Present
Robotic GeoAI for real-time obstacle detection
Bringing SPOT onto a Unitree Go2 Pro with an RGB-depth sensing camera, 360-degree LiDAR, and NVIDIA Jetson Orin Nano. The system combines detection, distance estimates, SLAM, map context, and spoken alerts to monitor obstacles along pedestrian routes.
2025–Present
Hardware-aware LLM inference and sustainable AI
I contribute to benchmarking the energy, water, and carbon footprint of LLM inference, studying how hardware and deployment choices shape latency, throughput, cost, and environmental impact. I extend this work through hands-on calibration of open-source models.
2024–2025
Context-aware RAG for mobile sensor anomalies
A framework that makes sensor anomalies interpretable and actionable. Using the ExtraSensory dataset, FAISS retrieval, sensor-pattern embeddings, and LLM reasoning, CLARA retrieves historical context to explain anomalies in mobile device sensor streams.
2024–2025
Multimodal sidewalk infrastructure assessment
Fusing drone imagery, flight metadata, and aerial LiDAR to scale sidewalk accessibility assessment. YOLO-OBB detections of crosswalks and truncated domes become real-world slope and dimension checks for ADA-PROWAG compliance.
2023–2024
Automated sidewalk crack detection and segmentation
Detecting cracks on rough, noisy, and overgrown sidewalks with RGB-depth fusion from an Intel RealSense camera. The pipeline combines DoT crack definitions, HSV filtering, automated labeling, and transformer-based segmentation.
2023–2024
Analyzing unlabeled urban sensor data with K-Means, DBSCAN, and HDBSCAN, then comparing temperature clusters with NOAA weather records to identify potentially faulty or unusual readings.
2020–2021 · M.S. thesis
An end-to-end system for Reddit text processing, tokenization, exploratory analysis, profanity detection, censorship, and n-gram language modeling, including practical edge cases in informal language.
2018–2019 · Honors thesis
Wrangling and visualizing 2016–2018 breach reports in R to understand changing attack patterns and practical countermeasures. Presented at MassURC and NCUR.
Preserving the details in object detection
Preserving weak but meaningful object features before decoder refinement to improve boundaries and separate adjacent instances.
SPOT: Spatially Preserving Object Detection with Transformers.
Koh, C. Y.; Hendawi, A.
Accepted to ACM SIGSPATIAL Research Track.
Assessing sidewalks from above
Combining drone imagery and aerial LiDAR to assess sidewalk slope, dimensions, and accessibility.
BirdCV-LiDAR: A Multimodal Data Fusion Framework for Automated Sidewalk Infrastructure Assessment.
Koh, C. Y.; Abdelatti, M.; Hendawi, A.
ISPRS Annals, XI–4–2026, 303–312.
Measuring the footprint of LLM inference
Benchmarking how hardware and deployment choices affect the energy, water, and carbon footprint of LLM inference.
How Hungry is AI? Benchmarking Energy, Water, and Carbon Footprint of LLM Inference.
Jegham, N.; Abdelatti, M.; Koh, C. Y.; Elmoubarki, L.; Hendawi, A.
Communications of the ACM (CACM), accepted for publication. CACM–25–11–5620.R2.
Detecting cracks in real sidewalk conditions
Detecting and segmenting sidewalk cracks with RGB-depth sensing, automated labeling, and transformer-based models.
CrackLens: Automated Sidewalk Crack Detection and Segmentation.
Koh, C. Y.; Ali, M.; Hendawi, A.
IEEE Transactions on Artificial Intelligence, 5(11), pp. 5418–5430.
ISPRS Annals, XI–4–2026, 303–312.
Accepted to ACM SIGSPATIAL Research Track.
Accepted to ACM SIGSPATIAL Demo Track.
Communications of the ACM (CACM), accepted for publication. CACM–25–11–5620.R2.
26th IEEE International Conference on Mobile Data Management (MDM), pp. 1–8.
Preprint · arXiv:2411.00201v4.
IEEE Transactions on Artificial Intelligence, 5(11), pp. 5418–5430.
Briefings in Bioinformatics, 25(3), bbae224.
Future Internet, 16(1), 28.
Proceedings of the 31st ACM International Conference on Advances in Geographic Information Systems, pp. 1–2.
Undergraduate Honors Thesis, Bridgewater State University.
Summers 2022, 2023
University of Rhode Island
Taught Survey of Computer Science to beginning students across majors, using clear examples, structured practice, and patient troubleshooting to make abstract ideas approachable.
Syllabus2021–2022, 2025–2026
University of Rhode Island
Supported undergraduate computing students through office hours, grading, and individual programming help, with an emphasis on debugging and independent learning.
Jan.–May 2019
Bridgewater State University
Managed tutor appointments and tutored business and economics calculus, advanced mathematics, and real analysis, balancing rigor with encouragement.
Sept. 2016–Dec. 2018
Bridgewater State University
Assisted students and taught a designated weekly class session, breaking algebraic and trigonometric concepts into approachable, step-by-step reasoning.
Summer 2018
Bridgewater State University
Provided individual support for students in mathematics, business, economics, and psychology, adapting examples to their fields and building confidence for future coursework.
Expected Fall 2026
University of Rhode Island
Kingston, Rhode Island
May 2021
Bridgewater State University
Bridgewater, Massachusetts
May 2019
Bridgewater State University
Departmental Honors · Minor in Mathematics
For conversations about research, teaching, GeoAI, accessible infrastructure, or sustainable AI, reach me by email.
Email me ↗