Projects

Ad Banner generation

• Led the development of an vision model trained in‑house to empower advertisers in generating visually stunning product backgrounds, significantly boosting Ad creativity.
• Optimized Qwen/SDXL ControlNet models with custom architecture to ensure product integrity, achieving scores on par with third‑party solutions like BRIA and Google Vertex AI.

E-KYC of Japanese Identification Cards for Rakuten Mobile

• Led E-KYC model development and deployment for Japanese ID cards, contributing to saving and platform cost reduction.
• Received GDSD Key Achievement and MEP Awards (Oct’21) for this contributions.

Pet insurance Banner generation And Rakuten Optimism booth

• Generated photorealistic images (dog breeds, car models) for insurance marketing banners, significantly boosting overall Click (2.2x for Dogs) and Conversion Rates.
• Developed an interactive AI system to generate children’s drawing-style illustrations using LoRA-based finetuning, based on keywords by children at the Rakuten Optimism event.

Image Translation

• Outperformed Google Vision and Paddle OCR with an in-house trained Japanese OCR model for seamless Rakuten Ad banner localization.

Clean Image Project

• Awarded Rakuten’s Most Impressive Project (Oct 2020) for the Clean Image Project. Trained a classification model to improve Rakuten Ichiba product image quality.
• Trained and fine-tuned ResNet models to classify product images as Pure/Not Pure, ensuring adherence to Rakuten Ichiba’s quality guidelines and enhancing customer trust.