Seattle, Washington, United States
Inference Optimization for Multimodal Understanding and Generation Models on Trainium Chips at Annapurna Labs
• Co-first-authored paper: A Multi-scalar and Multi-modal Approach to Architectural Heritage Documentation: An Interactive Digital Representation of the St. Nicholas Chapel. Reconstructed 3D indoor scenes in NeRF and Gaussian Splatting; captured and processed point cloud data from LiDAR and RGB-D camera for real-time 3D reconstruction using SLAM. Designed and prototyped an interactive document showcasing a multi-modal and multi-scalar approach to architectural heritage. • Processed 80k pixel-based floor plan images to extract vector and graph information; trained SOTA diffusion model on graph-conditioned floor plan image generation