Vanshil Shah

I am a Deep Learning Algorithm Engineer at NVIDIA, where I build and ship production inference microservices for large language models and vision-language models. I hold an M.S.E. in Robotics from the University of Pennsylvania and a B.E. in Mechanical Engineering from Nirma University, India.

My interests lie in large language models, deep learning, and high-performance inference — spanning custom backend optimizations in vLLM and TensorRT-LLM and end-to-end serving pipelines.

Previously, I worked on 3D reconstruction and computer vision at Quidient, autonomous vehicle perception at Ford, and SLAM research at the Indian Institute of Science.

Feel free to check out my Resume and drop me an e-mail if you want to chat with me!

Email  /  Resume  /  Linkedin  /  Twitter  /  Github

Shipped 10+ production inference microservices for large language and vision-language models to enterprise customers, implementing custom backend changes in vLLM and TensorRT-LLM.

Developed core Generalized Scene Reconstruction (GSR) algorithms for Quidient Reality®, producing relightable, millimeter-accurate 5D models from smartphone video. Built CUDA-enabled light physics module achieving 6x speedup in reconstruction.

Created a neural radiance field based model for augmenting datasets for downstream CV tasks

Deployed navigation stack on tracked robots used by Defence forces

Reconstructing dynamic LiDAR scans to static for improved SLAM performance

Collision avoidance framework accounting uncertainty of agent states without explicit model

Worked on enabling drone autonomy in smoke occluded environments for MBZIRC 2020

A complete library for Franka Panda arm was developed for manipulation

Implementation of Particle filter based SLAM algorithm for THOR humanoid robot

Implementation of stereo reconstruction and multi view stereo (plane sweep) algorithm from scratch

RC Snake Robot with serpentine gait controller

Website adapted from Jon Barron .