I build tools and pipelines for sim-ready and sim-optimized assets used in robotics simulation and synthetic data generation. I am particularly driven by opportunities to eliminate repetitive manual work through robust, scalable systems. Much of my work sits behind NDAs and standard internal processes, though I am always happy to discuss it at a high level. Outside work, I love motorcycling, following F1, cricket and MotoGP. I also build free indie games as a side quest. Here's one I'm currently working on, and this is for the geeks.
Current Work Experience
Amazon RoboticsTech Art SDE
Aug 2024–Present
Asset Pipeline:
Own the mesh processing layer of the CAD-to-USD pipeline used across several teams at AR.
Built an automated collider generation system supporting bounding shapes, optimal capsules for humanoid movement, convex hull & decomposition, optimal base shapes for SDF computation, and sphere approximation for robot arms. This eliminates any runtime overhead for collider computation & ensures all shapes are standardized across engines and validated beforehand.
Developed a novel parametric primitive decomposition pipeline that outperforms existing methods especially on hard-surface assets, saving ~1–2 hours of manual collider authoring per asset across 30k+ assets. Now exploring direct decomposition from BRep for lossless generation of shape parameters.
The pipeline ensures clean LOD workflows & assets are sim-ready and sim-optimized for a specific engine (e.g., Drake needs <15k poly count).
Established partnership with RapidPipeline & built a pipeline to integrate their tools org-wide, along with a suite of tools that simplify authoring context-dependent manual colliders needed for high-fidelity collisions.
Synthetic Data Generation:
Rearchitected the SDG framework into a self-serviceable, containerized, engine-agnostic pipeline by decoupling randomization, simulation, and rendering, enabling science teams to independently generate 1M+ datasets.
Built a GPU-accelerated writer library for niche scientific requirements such as damage maps and amodal segmentation, etc. using NVIDIA Warp.
Worked on optimizations that cut rendering compute costs by 4x and saving ~6 years in compute time.
Identified several bugs across Houdini & Isaac Sim, resolved some independently & worked with SideFX & NVIDIA to resolve the rest.
Developed a pipeline for PCG of 3D humanoids with varied topology, poses, clothes, props & features.
Previous Work Experience
CSAIL Lab, MITResearch Fellow
Jul 2024–Aug 2024
Selected for SGI 2024 cohort, where I worked on efficient robot designs for Lightspeed Studios. Also developed a novel VDF-based approach for extracting explicit meshes from implicit SDFs, addressing limitations of existing approaches.
Lockheed MartinTechnical Artist & Project Lead
Jan 2024–Apr 2024
Contracted with a team of 8 people and led them to develop a VR experience that demonstrates Lockheed's JADO system and has a modular 3D asset gallery with a conversational AI companion.
CrazyLabsGame Developer
Aug 2021–Aug 2022
Contracted as a third-party game studio and led a team of four to deliver 6 prototypes, 30 concept pitches, and an unannounced title, reaching a CPI (cost-per-install) as low as $0.28.
Education
FIEA, University of Central Florida
MS, Technical Art Track
2023–2024
GPA: 4.0/4.0
IIT Gandhinagar
B.Tech, Mechanical Engineering, Design & CSE Minor