Ao Kittri

I’m a software engineer in a computer-vision R&D lab, focused on large-scale 3D training-data pipelines for 3D generation and world-model research. I’ve built end-to-end workflows for generation, augmentation, quality validation, and model evaluation, with a strong bias toward catching data issues before training wastes compute. At HuaWei OD — R&D 3DV Lab, I independently traced a data-quality failure impacting ~200K of 1.2M meshes and built the filtering workflow to exclude the affected assets before a training run.

I also own recurring 3D data-processing and quality-control operations, from Python and Blender/bpy through Linux and SQL, and curate evaluation sets for generative-3D and avatar models. I orchestrated a pipeline that added ~210K half-body human scans by detecting facial keypoints on RGB renders with SCRFD and back-projecting them into 3D, then segmenting full-body scans to reduce the ~20% upside-down-crop rate from a prior bounding-box heuristic. To scale verification, I deployed Qwen3-VL-32B-Instruct across multi-node A100/H100 GPU clusters to detect 3D meshes carrying embedded text from their renders, processing 1M 3D assets and 16M renders.