Research scientist working on physical AI: sim-to-real transfer, digital twins, robot learning, vision-language models, edge AI, and medical robotics.
Summary
- Research scientist in physical AI with 8+ years of experience building AI systems that perceive, simulate, and act in the physical world, for real-world, resource-constrained settings where latency and reliability matter.
- I turn modern AI models into practical systems: sim-to-real transfer with synthetic data and city-scale digital twins, robot learning in Isaac Lab, vision-language and vision-language-action models, LLM post-training, large-scale data collection and annotation, edge inference, object detection, trajectory forecasting, and deployable research software.
Selected Contributions
- Physical AI and sim-to-real: Built Valdrada, a real-time digital twin of New York City that labels every frame with segmentation, depth, and 3D boxes (258 GitHub stars in six weeks), and G1 Patrol, humanoid navigation through crowded sidewalks trained in Isaac Lab. Boundless synthetic data outperforms CARLA renders for training detectors on real cameras, a sim-to-real gap measured on Constellation (IJCV 2026).
- Fielded AI systems: Leading multimodal perception projects for sensing models intended to scale across 900+ New York City intersections and CS3's three urban testbeds: COSMOS PAWR in New York City, DataCity in New Brunswick, and FAU MobIntel in West Palm Beach.
- Work at leading venues: Recent work in CVPR, ICML, ACM UIST, ACM/IEEE SEC, IEEE INFOCOM, IEEE PerCom, EDM, IJCV, eLife, and Surgical Endoscopy.
- Publication and systems record: 30+ public publications, preprints, software systems, and datasets spanning urban AI, medical robotics, neuroscience, and education.
- Open-source systems: Built public AI systems including DART and generative-agents with 1,700+ combined GitHub stars.
- Sponsored research: Proposal development, technical reporting, sponsor reviews, and engineering delivery across approximately $30M in institutional research supported by NSF, DARPA, AFOSR, and Con Edison, with additional project support from NVIDIA and EmpireAI.
- Teaching and mentorship: Taught Columbia graduate deep learning courses, mentored Master's, undergraduate, and high school researchers, six of whom went on to Ph.D. programs, and served as the main engineering instructor for the CS3 Research Experience for Teachers in 2024-2026.
Experience
2024–Present
Associate Research Scientist
- Lead large-scale applied ML projects in physical AI and robot learning, real-time urban perception, multimodal sensing, edge AI, vision-language systems, and medical robotics, including surgical world models.
- Created DART, a real-time open-vocabulary detector with TensorRT deployment and 790+ GitHub stars; led UrbanOmniDetect, an oral paper at the CVPR 2026 DriveX Workshop with public code, models, and data.
- {"Lead work toward a civic robot system for CS3"=>"Unitree G1 humanoid and B2-W quadruped navigation in dense pedestrian environments, trained in Isaac Lab with PPO teacher-student distillation, domain randomization, and a ROS 2 deployment stack."}
- Mentored the student first authors of papers at ICML 2025, IEEE INFOCOM 2026, and USENIX VehicleSec 2026, and released public work on distributed VLMs, real-time edge analytics, and medical robotics foundation model datasets.
- Designed and led the engineering curriculum for the CS3 Research Experience for Teachers in 2024, 2025, and 2026.
- Secured two NVIDIA research grants, with Zoran Kostic in 2025 and Andrew Smyth in 2026, and EmpireAI project support; INRIX x MetroLab Challenge finalist.
2022–2024
Postdoctoral Research Scientist
- Built computer vision systems and digital twins for urban intersections, including object detection, multi-object tracking, trajectory forecasting, and safety warning workflows.
- Applied computer vision to robotic surgery and endoscopy training with Northwell Health collaborators, yielding journal publications and conference presentations.
- Mentored 11 Master's students across deep learning, computer vision, and applied AI research projects.
- Received the Columbia University Electrical Engineering Distinguished Teaching Award and a National Postdoc Appreciation Week Excellence Award.
2023
Lecturer in Deep Learning
- Designed and taught Neural Networks & Deep Learning and Advanced Deep Learning to 123 students, covering computer vision, transformers, diffusion models, and modern generative AI.
2017–2022
Ph.D. Researcher
- Built FlyBrainLab from scratch as an early AI workbench for research, combining free-form scientific querying with literature and ontology retrieval, a connectome knowledge graph in OrientDB, large-scale executable queries, GPU-accelerated neural simulation, and interactive 3D visualization through a TypeScript/JupyterLab interface.
- Published connectome-scale computational neuroscience work in eLife and Frontiers in Neuroinformatics.
Sponsored Research Contributions
2022–Present
NSF Engineering Research Center for Smart Streetscapes (CS3)
- Coordinate applied AI research, annual reporting, and cross-team demonstrations across a five-institution center; present DART, UrbanOmniDetect, and related systems at annual NSF site visits.
2021–Present
NSF CPS: Hybrid Twins for Urban Transportation
- Led engineering delivery, prepared annual technical reports, and presented project reviews to NSF for traffic sensing, prediction, and digital twin systems spanning individual intersections and citywide operations.
2019–2020
DARPA Robust Learning in Brain Circuits of Synthetic Miniature Insects
- Contributed to proposal development, authored project reports, and conducted grant-supported doctoral research on robust learning and executable brain circuit models.
2017–2021
AFOSR Foundations of Neuroinformation Processing
- Developed computational systems and models for phase- and spike-based neural information processing, including work underlying the FlyBrainLab platform.
2017–2019
NSF Collaborative Research: The Digital Fly Brain
- Engineered connectome data, graph query, simulation, and visualization capabilities for the open-source Fruit Fly Brain Observatory and FlyBrainLab systems.
2015–2016
Con Edison Underground Infrastructure and Public Safety Analytics
- Graduate Researcher: Developed theoretical models, implementations, and data processing pipelines for a Visual Data Capture pilot using thermal imagery, high-resolution photographs, and video of underground electrical structures. The work examined automated hotspot detection, inspection targeting, and estimation of avoided serious events.
- Extended the analysis system with infrared camera data assimilation, equipment and cable density models, temperature differential extraction along cable lengths, and visual features associated with later failures; delivered project outputs to Con Edison.
- Contributed cost-benefit and portfolio optimization analysis for Con Edison's secondary public safety programs.
Awards and Recognition
2026
NVIDIA Research Grant
2026
Best Storytelling Runner-Up, MetaMorph AI Award
2025
Best Paper Award, ACM/IEEE Symposium on Edge Computing
2025
NVIDIA Research Grant
2025
INRIX x MetroLab Challenge Finalist
2024
Columbia University Electrical Engineering Distinguished Teaching Award
2023
Smart Cities North America Awards Winner, Transportation
2023
National Postdoc Appreciation Week Excellence Award
Selected Publications and Public Systems
- Constellation Dataset: Benchmarking High-Altitude Object Detection for an Urban Intersection. International Journal of Computer Vision 134(10), 2026.
- Calibration-Free View-Agnostic Monocular 3D Object Detection for Urban Scenes. CVPR Workshops 2026.
- AI-Powered CPS-Enabled Vulnerable-User-Aware Urban Transportation Digital Twin: Methods and Applications. IEEE Transactions on Intelligent Transportation Systems 27(5), 2026.
- Worst-Case Attacks in Reactive Edge-Cloud Systems: Expected Latency and SLA Violation. IEEE INFOCOM 2026.
- Digital Eyes on the Road: Using Street Cameras to Verify Traffic Integrity and Detect Sybil Attacks at City Scale. USENIX VehicleSec 2026.
- Detect Anything in Real Time: From Single-Prompt Segmentation to Multi-Class Detection. DART, arXiv 2026. Public code and models.
- Valdrada. A real-time digital twin of New York City compiled from public records, with segmentation, depth, and 3D box labels for every frame and a Python API modeled on CARLA, 2026. Public code and data.
- G1 Patrol: Humanoid Navigation in Dense Urban Pedestrian Environments. Training and evaluation in Isaac Lab for the Unitree G1, 2026. Public code.
- Open-H-Embodiment: A Large-Scale Dataset for Enabling Foundation Models in Medical Robotics. arXiv 2026.
- Harnessing Floating Car Data, Traffic Camera Observations, and Network Flow Analysis for Traffic Volume Estimation. Discover Civil Engineering 3, 257, 2026.
- Newell-Car-Following-Model-Informed Gaussian Process Regression Model for Ultra-Short-Term Arterial Vehicle Trajectory Prediction. Transportation Research Record 2026, to appear.
- Adapt, Label, Distill: Foundation-Model Supervision for Efficient Object Detection. WACV 2027 Real-World Surveillance Workshop, to appear.
- Real-Time Video Analytics for Urban Safety: Deployment over Edge and End Devices. ACM/IEEE SEC 2025, Best Paper Award.
- Boundless. Unreal Engine 5-based photorealistic synthetic data generation for object detection in urban streetscapes, arXiv 2024.
- Adaptive Data Collection for Robust Learning Across Multiple Distributions. ICML 2025.
- Distributed VLMs: Efficient Vision-Language Processing through Cloud-Edge Collaboration. IEEE PerCom Workshops 2025.
- StreetNav: Leveraging Street Cameras to Support Precise Outdoor Navigation for Blind Pedestrians. ACM UIST 2024.
- FlyBrainLab: Accelerating the Discovery of the Functional Logic of the Drosophila Brain. eLife 2021.
Technical Skills
AI and Machine Learning
- Computer vision
- Object detection, segmentation, tracking, monocular 3D detection, video analytics
- Generative and multimodal AI
- VLMs, VLAs, diffusion models, world models, RAG, agents, LLM fine-tuning and post-training (DPO, RL)
- Data-centric AI
- Large-scale collection, annotation (CVAT), synthetic data, benchmarks, adaptive sampling
- Knowledge and graph systems
- Ontology retrieval, OrientDB/NeuroArch graph databases, large-scale graph querying
Simulation and Robotics
- Robot learning
- NVIDIA Isaac Lab and Isaac Sim, Newton (MuJoCo-Warp), ROS 2, PPO, teacher-student distillation, domain randomization, sim-to-real transfer, Unitree G1 humanoid and B2-W quadruped navigation from lidar and depth sensing
- Simulation
- Unreal Engine 5 (UE5), CARLA, OpenUSD, Blender, city-scale digital twins, traffic and pedestrian simulation
- Synthetic data and testing
- Photorealistic synthetic data with labels computed from the simulation, scenario-based and hardware-in-the-loop testing
Systems and Deployment
- Edge inference
- TensorRT, ONNX, LiteRT, NCNN, WebGPU, NVIDIA Jetson, Qualcomm, Raspberry Pi, on-device LLMs, edge and cloud collaboration, low-latency pipelines
- Frameworks
- PyTorch, JAX, TensorFlow, Hugging Face, Ultralytics, OpenCV, CUDA
- Languages
- Python, C/C++, CUDA, TypeScript, JavaScript, Node.js, SQL
- Data and visualization
- OpenStreetMap, NYC Open Data, Three.js/WebGL, scientific visualization
Product and Communication
- Technical leadership
- Cross-institution projects, mentoring, curriculum design, public demos
- Design and media
- Visual storytelling, Figma, Photoshop, Illustrator, InDesign, Premiere
Education
2022
Ph.D. in Electrical Engineering
- Research area: Systems Biology and Neuroengineering
- GPA: 4.10/4.33
- Herbert French Fellowship; Helmsley Fellowship for the Cold Spring Harbor Laboratory Drosophila Neurobiology course
2016
M.Sc. in Computer Science
- Machine learning and thesis track
2015
B.Sc. in Electronics and Communication Engineering
Leadership and Teaching
2024–2026
Main Engineering Instructor, Research Experience for Teachers
- Designed and led the core engineering curriculum for three annual cohorts, covering neural networks, model evaluation, computer vision, annotation, YOLO training, object tracking, generative AI, and edge deployment.
- Supported K-12 educators in developing reproducible, classroom-ready lesson plans that connect AI engineering to science, mathematics, civics, and career education.
2024–2025
Co-Chair, CS3 Student Leadership Council
2023
Lecturer, ECBM E4040 Neural Networks and Deep Learning
2023
Lecturer, EECS E6691 Advanced Deep Learning
Selected Presentations
Sep 2026
A Vision-Based Analysis of Congestion Pricing in New York City
May 2026
Center for Smart Streetscapes NSF Annual Site Visit
Feb 2026
From Sensors to Systems: Real-Time Traffic Analysis for Faster Decision-Making
Jan 2026
Port Authority Tomorrow Summit Workshop
Nov 2025
Evaluating Micromobility & Dangerous Riding Behaviors
Nov 2020
FlyBrainLab - An Interactive Open Computing Platform
Professional Service
- Organizing committees: Poster and Demo Co-Chair, IEEE SMARTCOMP 2027 (13th IEEE International Conference on Smart Computing), Boca Raton, Florida.
- Conference reviewing: CVPR 2026, ICML 2026, ICMLA 2026, WACV 2025, and ACM MobiCom 2024.
- Journal reviewing: IEEE/ACM Transactions on Networking, IEEE Transactions on Very Large Scale Integration Systems, Sensors (MDPI), Electronics (MDPI), and Smart Cities (MDPI).