cv

Robotics and machine learning engineer. Currently training and deploying Vision Language Models at NVIDIA.

General Information

Full Name Naman Antony Menezes
Location Santa Clara, CA
Email namanamenezes@gmail.com

Education

  • 2023 - 2025
    M.S. in Robotics Systems Development
    Carnegie Mellon University, School of Computer Science
    • GPA 4.06 / 4.00
  • 2018 - 2022
    B.E. in Electronics and Communication
    R. V. College of Engineering
    • GPA 9.14 / 10.00

Experience

  • 2025 - now
    Software Engineer
    NVIDIA Corporation, California
    • Trained a 600M-parameter Vision Language Model (DINOv2 encoder, Llama 3.1 backbone) with PyTorch DDP on 64 H100 GPUs across 8 nodes, followed by GRPO reinforcement learning post-training.
    • Shipped the improved MapGPT to production, delivering HD maps across multiple US cities and labelling the entire San Francisco area in hours.
    • Built the production inference stack on vLLM with caching and download optimizations, scaling VLM inference across AWS Kubernetes GPU clusters and driving thousands of GPU-hours.
    • Cut map generation cost ~5x by running inference only along road geometry instead of full tile grids.
    • Improved output precision and recall by 10% through inference-time scaling.
  • 2024
    Software Engineering Intern
    NVIDIA Corporation, California
    • Designed an exporter converting 3D Occupancy Pointcloud Maps from NVIDIA's proprietary format to the open source 3D Tile Format, with geospatially aligned tiles at multiple levels of detail.
    • Scaled the architecture to run on massively parallel cloud servers, exporting city scale maps in under 5 hours.
    • Implemented meshoptimizer compression, reducing 3D Tile Map size by ~5x and enabling city-scale visualization over a browser for the first time.
  • 2022 - 2023
    Project Associate
    Artificial Intelligence and Robotics Lab (AIRL), Indian Institute of Science, Bangalore
    • Developed an Autonomous Driving Stack for GPS Denied Environments, presented at CES 2024.
    • Built the entire low level communication architecture over CAN Bus.
  • 2022
    Software Engineering Intern
    Packet Forwarding Engine Team, Juniper Networks, Bangalore
    • Developed a Python tool that drastically reduces time spent debugging routers.

Projects and Competitions

  • 2026
    RoboHacks, Y Combinator
    • Winner, Overall Prize and Scale AI Data Track ($2,500 and two Innate robots).
    • Built cross-robot spatial memory turning iPhone video and narration into a shared 3D map for pick and place.
    • Integrated NVIDIA Parakeet ASR, 3D reconstruction, ICP, Gemma 3 VLM and an action chunking policy.
  • 2024
    Meta Reality Labs: Photorealistic Human Avatars using Drones
    • Created a human following drone that avoids obstacles using the DJI SDK and ROS2 on a Mavic 3 Enterprise.
    • Trained a model on synthetically generated avatars and demonstrated a real time full body codec avatar.
  • 2023
    GPS Denied Autonomous Navigation
    • Coded a full stack autonomous navigation algorithm for GPS denied environments.
    • Developed LiDAR map to real world registration using a particle filter, leveraging the PCL library.
    • Implemented local and global planning giving real time obstacle avoidance and road alignment.
  • 2023
    2D LiDAR Only SLAM
    • Designed a CUDA based 2D LiDAR SLAM algorithm in C++ for ROS1, optimized to run in real time.
    • Implemented a Fourier Mellin Transform frontend and an SE-Sync (graphSLAM) backend.

Technical Skills

  • Languages
    • Python, C/C++, CUDA, MATLAB, Verilog HDL
  • Machine Learning
    • PyTorch, vLLM, Vision Language Models, DDP, GRPO, reinforcement learning
    • Distributed multi-node training, fine-tuning, inference optimization
  • Robotics
    • ROS1/ROS2, Gazebo, SLAM, sensor fusion, motion planning, PCL, LiDAR, point clouds, CAN Bus
  • Infrastructure
    • AWS, Kubernetes, Docker, Linux, Git, CMake

Other Interests

  • Hobbies: Basketball, piano, foosball.