Meta FAIR Labs (via Insight Global) – Technical Artist

Published Research: PARTNR: A Benchmark for Planning and Reasoning in Embodied Multi-Agent Tasks
(co-author)

For two years I worked on HSSD-Hab, a dataset enhancing Habitat 3, Meta’s open-source simulation platform for embodied AI. My focus was scaling 3D asset pipelines, automating QA, and improving visual fidelity to support large-scale AI training.

Key Contributions

Asset QA & Automation:

Built custom Blender/Python tools to validate and export 18,000+ rigid assets and 2,000+ articulated objects, reducing iteration time by 50%.

Automated collider corrections, material fixes, and URDF exports, condensing multi-step processes into single-click workflows.

Scene & Environment Enhancements:

Designed a region annotation system for 211 scenes, refining navigation, spatial semantics, and object interaction.

Upgraded environments with PBR materials; automated deployment pipelines improved consistency and cut manual workload.

Process Optimization:

Standardized QA tracking for thousands of assets using Google Sheets + Python automation.

Reduced validation errors by 90%, improved dataset scalability, and doubled processing efficiency.

Impact

My work enabled Habitat 3 to handle vast 3D datasets more efficiently, with higher visual realism and consistency. These contributions strengthened AI research pipelines while demonstrating expertise in:

Large-scale asset management & CAD-like optimization

Tool development in Blender/Python for automation

Unity-adjacent workflows for interactive simulations

Pipeline scalability for real-time 3D environments