Computational Photography Lab, advised by Prof. Yu-Lun Liu.

◦ Co-first authored AuraFusion360, published at CVPR 2025 in collaboration with NVIDIA Research team and Prof. Yu-Lun Liu.

◦ Developed a state-of-the-art 3D object removal method for 360° unbounded scenes, ensuring multi-view consistency and geometric plausibility.

◦ Leveraged depth-aware diffusion models with iterative fine-tuning, achieving +2 dB PSNR and –0.15 LPIPS improvement over prior state-of-the-art.

◦ Integrated AI chatbot workflow into customer service workflows, reducing human assistance by 55% and saving $500K annually.

◦ Migrated 20+ AWS Lambda functions to Kubernetes Jobs, lowering infrastructure costs and database expenses across 10+ countries.

◦ Reduced 45% cashback service incidents by identifying outdated Kafka data streams and introducing a distributed Redis lock to prevent incorrect updates.

◦ Built a B2B web application enabling partners to manage product portfolios and execute targeted

campaigns.

◦ Integrated third-party BNPL service, driving a 20% revenue increase within the first month.

◦ Collaborated with product managers, designers, and merchants to redesign shopping flow, improving usability and conversion rates.