Stanford
MS, Computer Science
GPA 4.17/4.0
Distinction in ResearchBuilding the machine that builds the machine.
I’m a researcher at OpenAI, working on frontier models including GPT-6, the GPT-5 series, and the o-series. My focus is on retrieval, organizational intelligence, and teaching models to perform useful work in private contexts. My work has spanned human action understanding in video, perception for autonomous vehicles, and large-scale AI models.
Previously, I was a tech lead in perception at Waymo, helping autonomous vehicles understand the most complex part of the road: people. I studied computer science at Stanford and IIT Kanpur, where my research focused on video understanding and learning from visual data with limited supervision.
In a prior life, I spent time on competitive programming and algorithmic trading.
Recent work
-
GPT-6 Astra
Reasoning & scientific work
-
GPT-5.6
Coding & professional work
-
GPT-5.5
Planning & tool use
-
GPT-5.4
Reasoning & computer use
-
GPT-5.2
Long-running agents
-
GPT-5
Unified reasoning
-
o3 & o4-mini
Visual reasoning & tools
-
Company knowledge
Answers from workplace sources
-
Demonstrably safe AI
Driving, simulation & safety validation
-
Onboard VLM
World knowledge meets autonomous driving
-
Human keypoints
Human movement & intent
-
Home Action Genome
Actions across views & modalities
-
CoCon
Multimodal video representations
Education & Recognition
IIT Kanpur
BTech, Computer Science
GPA 9.9/10.0
- 35 under 35: Self Driving
- Gold Medal: Best performance in batch
- IPhO training camp Top 30 nationally
- IMO training camp Top 6 nationally
- CS231nCS145 Teaching assistant
- ESC101 Section leader
Talks & service
- Talks
- BMVC
- Reviewing
- WACV; workshops at CVPR, ICCV, NeurIPS, ICRA & BMVC.
- Committees
- Workshops at CVPR, NeurIPS, ICRA & BMVC.
Publications
GPT-5 System Card
Capabilities, safety evaluations, and mitigations for the GPT-5 model family.
Home Action Genome: Cooperative Compositional Action Understanding
Learning compositional human actions from multiple views and modalities.
Paper
Project
Code
@inproceedings{rai2021home,
author = {Rai, Nishant and Chen, Haofeng and Ji, Jingwei and Desai, Rishi and Kozuka, Kazuki and Ishizaka, Shun and Adeli, Ehsan and Niebles, Juan Carlos},
title = {Home Action Genome: Cooperative Compositional Action Understanding},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
year = {2021},
pages = {11184--11193},
month = {June},
url = {https://arxiv.org/abs/2105.05226}
}
CoCon: Cooperative-Contrastive Learning
Self-supervised video representations from RGB, motion, and pose.
Paper
Code
@inproceedings{rai2021cocon,
author = {Rai, Nishant and Adeli, Ehsan and Lee, Kuan-Hui and Gaidon, Adrien and Niebles, Juan Carlos},
title = {CoCon: Cooperative-Contrastive Learning},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
year = {2021},
pages = {3384--3393},
month = {June},
url = {https://arxiv.org/abs/2104.14764}
}
Weak Multi-View Supervision for Surface Mapping Estimation
Learning 3D shape and surface correspondence through consistency across views, without dense annotations.
Paper
@inproceedings{rai2021surface,
author = {Rai, Nishant and Liaudanskas, Aidas and Rao, Srinivas and Cayon, Rodrigo Ortiz and Munaro, Matteo and Holzer, Stefan},
title = {Weak Multi-View Supervision for Surface Mapping Estimation},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
year = {2021},
pages = {2759--2768},
month = {June},
url = {https://arxiv.org/abs/2105.01388}
}
AdaScan: Adaptive Scan Pooling in Deep Convolutional Neural Networks for Human Action Recognition in Videos
Selecting informative frames for recognizing actions in video.
Paper
Project
Code
@inproceedings{kar2017adascan,
author = {Kar, Amlan and Rai, Nishant and Sikka, Karan and Sharma, Gaurav},
title = {AdaScan: Adaptive Scan Pooling in Deep Convolutional Neural Networks for Human Action Recognition in Videos},
booktitle = {Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year = {2017},
month = {July},
url = {https://arxiv.org/abs/1611.08240}
}
Partial Multi-View Clustering using Graph Regularized NMF
Clustering incomplete multi-view data using nonnegative matrix factorization and graph regularization.
Paper
Code
@inproceedings{rai2016partial,
author = {Rai, Nishant and Negi, Sumit and Chaudhury, Santanu and Deshmukh, Om},
title = {Partial Multi-View Clustering using Graph Regularized NMF},
booktitle = {2016 23rd International Conference on Pattern Recognition (ICPR)},
year = {2016},
pages = {2192--2197},
doi = {10.1109/ICPR.2016.7899961},
url = {https://doi.org/10.1109/ICPR.2016.7899961}
}