Applied ML Engineer | 3D Perception | Production Systems

Sumesh Thakur Mission-ready ML systems, built with research depth and production discipline.

I design and ship perception pipelines that work under real constraints—from point-cloud research to embedded deployment. Measurable outcomes, reliable systems, clear communication.

Explore Deployed Systems Full Manifest

GitHub | LinkedIn | Scholar | Email

5+ Years in ML & CV

3 Peer-Reviewed Works

3+ Production Systems Shipped

2 Domains: Research + Industry

Systems Deployed

Production perception systems with measurable impact

Featured Repository

PointCloudSampling

Utility library for uniform and Poisson-based downsampling, built for practical 3D preprocessing and rapid experimentation.

Faster iteration for perception data pipelines
Clear baseline comparisons for sampling quality
Reusable foundation for research and demo workflows

Open Repository

Falkbuilt

3D Mesh Reconstruction Pipeline

Built ETL and reconstruction workflows for architectural mesh assets using Python, MongoDB, and Trimesh.

Production CV pipeline

JCA Technologies

Embedded Perception Stack

Delivered LiDAR-camera fusion, SLAM, and visual odometry modules on NVIDIA Jetson and autonomy platforms.

Real-time deployment

BarrierBreak Solutions

6 by 6 Accessibility App

Led Android currency-detection deployment using TensorFlow Lite and large-scale dataset curation for robust edge inference.

Assistive AI

Research Depth

Peer-reviewed work in 3D learning and graph systems

IEEE IGARSS 2023

GAGAT: Global-Aware Graph Attention for 3D Classification and Segmentation

View Paper

IEEE IGARSS 2022

Graph Attention Network for Object Detection from Raw LiDAR Data

View Paper

arXiv 2020

Dynamic Edge Weights in GNNs for 3D Object Detection

View Preprint

View Full Scholar Profile

Interactive Lab

Hands-on implementations demonstrating technical depth

Cornell Box Path Tracer

Interactive Monte Carlo rendering demo with controllable sampling, depth, lighting, and camera parameters.

Launch Demo

Point Cloud Processing

Browser-based downsampling, segmentation, and clustering workflow with performance instrumentation.

Launch Demo

Procedural Terrain Generator

Realtime procedural 3D terrain synthesis with adjustable noise and camera controls.

Launch Demo

ESTABLISHING CONNECTION...

Open to New Missions

Available for ML engineer and perception roles

If your team needs someone who bridges rigorous research and production reliability, I'm available for conversation. Let's discuss how I can contribute to your mission.

View Resume Download Email

GitHub LinkedIn Google Scholar Resume