Impact
Recognition, coverage and public talks, open-source adoption, citations, teaching, and where my mentees went next.
Counts and positions are as of October 2026, and every item links to its original source.
Recognition
Best Storytelling Runner-Up, MetaMorph AI Film Award
Best Paper Award, ACM/IEEE Symposium on Edge Computing (SEC 2025)
Finalist, INRIX x MetroLab Challenge
Distinguished Teaching Award, Columbia Electrical Engineering
Smart Cities North America Award, Transportation, IDC Government Insights
National Postdoc Appreciation Week Excellence Award, Columbia University
Coverage and public talks
PAWR PI Meeting 2026
The meeting highlights feature my abstract on measuring the effect of congestion pricing from more than 900 public traffic cameras.
How AI Is Improving Daily Life in Cities
Coverage of the CS3 Innovation Summit panel, quoting my work on calibrating a citywide traffic model with camera data.
From Smart Streets to Smarter Cities
Written with Andrew Smyth, Antonina Kosikova, and Boshra Khalili on validating and scaling traffic volume estimation across New York City.
The Confidence Gap: How to Integrate AI and Traffic Data for Operational Decision-Making
An independent write-up of my talk on combining large-scale computer vision with physics-constrained graph neural networks to model traffic across New York City.
Panelist, From Sensors to Systems: Real-Time Traffic Analysis for Faster Decision-Making
With Robert Holbrook of the NYC Mayor's Office and Ahmed Darrat of INRIX.
Port Authority Tomorrow Summit Workshop
Research Team Showcases AI-based Urban Traffic Applications at CVPR 2024
A data-driven traffic simulation for metropolitan street intersections, presented with Chengbo Zang and the COSMOS team.
Used by others
Valdrada
- 140 GitHub stars in its first three weeks.
- Its streamed-tile architecture is followed by Procedural Tokyo and, through it, a San Juan port. 3DStreet documented it as prior art, and Databook.nyc asked to link it from its building pages.
UrbanOmniDetect
- 4,458 downloads of the released model on Hugging Face.
- Benchmarked at 30 frames per second on a laptop's integrated GPU by condados.ai, and ported to C++ and Qt as OmniBEV.
DART
- 781 GitHub stars and 93 forks in its first seven months; the launch announcement drew more than 2,500 reactions on LinkedIn.
- Proposed for integration into Intel's Open Edge Platform, with OpenVINO models, and into the OCTRON animal-tracking tool.
- Benchmarked against DeepLabCut at the Lemanic Life Sciences Hackathon 2026, extended by DAIRT to detect interactions between objects, and listed in Awesome Segment Anything.
NYC congestion pricing analysis
- The Center for Smart Streetscapes runs the project's real-time camera map on its own site.
generative-agents
- 994 GitHub stars and 163 forks.
- The base of the Chinese generative agents project in the Luotuo LLM community, and featured in the Korean Weekly AI ArXiv notes.
Citations
- 275 citations on Google Scholar and an h-index of 12, growing from 26 citations in 2024 to 77 in 2025 and 132 so far in 2026.
- The suturing world models paper is cited by NVIDIA’s Cosmos-H-Surgical, and DART by a survey of efficient Segment Anything models in the International Journal of Computer Vision.
- FlyBrainLab’s eLife paper has an Altmetric score higher than 87% of papers from the same period.
Teaching and outreach
Co-chair, CS3 Student Leadership Council
Lecturer, Columbia University
Neural Networks and Deep Learning and Advanced Deep Learning for 123 graduate students, recognized with the 2024 Distinguished Teaching Award.
Engineering mentor, CS3 My Streetscape
A high school research project on AI models for digital twins of street intersections.
Mentees
I mentored the following students in their research: Master’s students as the postdoctoral researcher in Columbia’s AIDL Lab, and undergraduate and high school researchers on FlyBrainLab during my Ph.D. in the Bionet Group.
Now in graduate school
- Jonathan Marty joined the Bionet Group as a high school intern in 2018, packaging the Fruit Fly Brain Observatory and modeling the fly’s early visual system, and continued on FlyBrainLab as a Columbia undergraduate. Ph.D. student in applied and computational mathematics at Princeton.
- Gyung Hyun (Jay) Je, Columbia MS in Computer Science, worked on the Constellation dataset (IJCV 2026). Ph.D. student at the University of Toronto.
- Haoyu Dong, Columbia MS, worked on robust LLM post-training (2026). Ph.D. student at UC Santa Cruz.
- William Ho, Columbia MS and Fulbright Scholar, worked on zero-shot image tagging at traffic intersections (ICMLA 2026). Ph.D. student in Columbia Electrical Engineering, with a Byron Fellowship.
- Chengbo Zang, Columbia MS in Electrical Engineering, worked on Constellation, Boundless, and adaptive data collection (ICML 2025, first author). Ph.D. student in Columbia Electrical Engineering.
- Yuyang (Ian) Li, Columbia MS in Electrical Engineering, worked on distributed vision-language models (IEEE PerCom Workshops 2025, first author) and Boundless. Ph.D. student in Columbia Electrical Engineering.
- Feiyang (Kathy) Yu, Columbia undergraduate in Computer Science, worked on computational models and simulation engines for the fly brain in the Bionet Group. Went on to graduate study in computer science at Stanford.
Now in industry
- Blake Castleman, Columbia MS in Computer Science, worked on GPT-based intelligent tutors (EDM 2024). Software engineer at Anduril.
- Sanjeev Narasimhan, Columbia MS in Computer Science, worked on Constellation and surgical video analysis. Applied scientist at Johnson & Johnson.
- Feroz Ahmad, Columbia graduate student, worked on surgical phase recognition (Bioengineering 2023). Machine learning engineer at Meta.
- Devika Gumaste, Columbia MS in Electrical Engineering and EE MS Award of Excellence recipient, worked on UrbanOmniDetect and distributed vision-language models. Works on multimodal AI at Known.