Working Trees

Built production ML pipelines that reconstruct 3D tree models from single images. Used reinforcement learning to fine-tune SAM and Depth Anything, optimized for lower-end mobile devices, and cut compute time by 50%. Developed original video-to-3D algorithms hitting 95%+ scale accuracy using NeRF and Gaussian Splatting.

Stanford StartX program, working directly with professors and environmental scientists to apply their published research.