How I Think
Most robotics companies treat perception as a preprocessing step — a box that spits out coordinates. I think that's exactly backwards. Perception is the bottleneck. It's the thing that determines whether your robot can handle the real world or only works in a cage.
The gap between a demo and a product in robotics is almost always a perception problem. You don't need better planners. You need eyes that actually work.
I've spent my career closing that gap: building foundation models that generalize to unseen objects, writing CUDA kernels that run in single-digit milliseconds, and shipping systems that work in production — not just in simulation.
Career
Gritt Robotics
Perception Lead
2024 – Present
Leading perception systems for autonomous robotic operations. Own the full vision stack.
Intrinsic (Alphabet)
Senior Software Engineer, Perception
2021 – 2024
- #1-ranked vision foundation model on the BOP benchmark for 6DoF pose estimation ICCV 2025
- Open-sourced 100K+ image dataset for pose estimation research, with a Real2Sim data engine using Gaussian splatting CVPR 2024
- 2 US patents granted (17/471,063 & 17/393,338) for pose estimation and rendering methods
- Demoed perception systems at major robotics expos Automate 2024 Automatica 2023
Akasha Imaging
Software Engineer, Computer Vision
2020 – 2021
Early engineer at a Bay Area deep imaging startup (MIT News). Built core vision pipelines. Acquired by Alphabet → became Intrinsic.
- Towards Rotation Invariance in Object Detection (ICCV, 2021)
EXL
EXL Services
Consultant, Analytics
2016 – 2018
Data science consulting for retail and media clients.
Open Source
Education
University of Waterloo
Master of Mathematics (MMath), Computer Science
2018 – 2020
Thesis in computer vision & object detection. Reviewer for CVPR, AAAI, BMVC, NeurIPS, PLOS ONE. Mitacs Accelerate Fellow. CIFAR DLRL Summer School 2020.
- Do More Stringent Policies Reduce Daily COVID-19 Case Counts? Evidence from Canadian Provinces (Economic Analysis and Policy, 2023)
- Sensitivity to Risk Profiles of Users When Developing AI Systems (LNCS/LNAI, 2020)
- PocketView: Through-Fabric Information Displays (UIST, 2021)
Indian Institute of Technology, Kanpur
B.Tech, Materials Science & Engineering
2012 – 2016
Research intern at Rutgers University & Academia Sinica, Taiwan. Published at UbiComp 2016 and PLOS ONE. Quiz Club coordinator.
Featured Work
Using Deep Imaging for Higher Resolution
MIT News, 2022
Computer Science Researchers Developing Through-Fabric Display Technology
University of Waterloo, 2021
Elsewhere
- Poetry — I write poems, sometimes about code, mostly not
- Photography — exhibited at 5th Mega Heritage Photo Exhibition, India Habitat Center, New Delhi (2016). Featured in iPhoneLife Magazine and "100 Phone Photographers to Follow on Instagram"
- 200dishes — cooked one dish from every country on earth
Advising & Investing
I occasionally advise and angel invest in early-stage robotics and applied ML companies. If you're building something interesting, reach out.
Contact