Chris Ku

Senior Machine Learning Engineer

Raspberry AI

Feb 2024 - Current (2 years 5 months)

• Developed and deployed production generative AI models for fashion design automation, enabling enterprise clients to transform sketches into photorealistic product renderings. • Architected a conversational AI design agent using MCP and Google Vertex AI with multi-intent classification, enabling natural language image generation with contextual reference image resolution from conversation history. • Built custom LoRA fine-tuning pipelines for brand-specific model personalization, delivering customized AI design tools for Fortune 500 apparel and footwear companies. • Designed and implemented a semantic asset retrieval system using Fashion-CLIP embeddings, enabling sub-second similarity search across generated fashion assets. • Engineered t

Senior Machine Learning Software Engineer

Google

Jun 2020 - Dec 2023 (3 years 7 months)

• Co-developed MobileNetV4, a state-of-the-art efficient neural network architecture achieving 87% ImageNet accuracy at 3.8ms inference latency on Pixel 8 EdgeTPU. • Served as core developer on TensorFlow Model Garden, implementing and deploying vision models supporting Cloud, Chrome, Photos, AR, and Pixel products used by over 1 billion users worldwide. • Designed and implemented Quantization-Aware Training (QAT) support for Model Garden, enabling INT8 deployment of MobileNet and ResNet families with minimal accuracy loss and 3x latency reduction. • Led multimodal model fine-tuning research for Google Gemini, developing techniques for interleaved text and image data processing to enable diverse product use cases across Google ecosystem. •

Senior Data Engineer

Valtech

Aug 2017 - May 2020 (2 years 10 months)

• Designed and maintained large-scale ETL pipelines using Python, Spark, and Airflow, automating ingestion and transformation of multi-terabyte datasets from digital platforms into AWS Redshift and Snowflake. • Partnered with data scientists and product teams to build predictive analytics and real-time personalization systems, driving customer engagement optimization for enterprise e-commerce and retail clients. • Developed and optimized dimensional data models and business metrics, improving query performance and reducing reporting latency by 40% across client analytics platforms. • Implemented CI/CD workflows with GitLab and Terraform, enabling automated deployment and versioned data pipelines across staging and production environments fo

Research Scientist

ebay

Jun 2014 - Jun 2017 (3 years 1 month)

• Architected and deployed large-scale visual search systems serving 1 billion+ live product listings, implementing deep learning models for image classification and similarity matching. • Created ModaNet, a large-scale street fashion dataset with 55,176 polygon-annotated images across 13 product categories, published as oral presentation at ACM Multimedia 2018 and adopted as industry benchmark. • Developed fine-grained visual classification models for fashion product understanding, enabling automated categorization and attribute extraction for millions of seller listings. • Implemented efficient image indexing and hashing algorithms for real-time visual search at scale, optimizing retrieval latency and accuracy across eBay's product catalo

Research Assistant

UNIVERSITY OF WISCONSIN-MADISON

Jan 2013 - Jan 2014 (1 year 1 month)

• Developed efficient object detection algorithms using deep CNNs for autonomous driving applications, achieving best detection accuracy on internal benchmark datasets. • Designed regression-based similarity fusion algorithms for real-time image retrieval systems, significantly improving retrieval accuracy and speed for logo recognition. • Implemented hashing algorithms and CNNs for scalable video-based face verification, enabling efficient biometric matching across large video databases. • Built multi-modal fusion algorithms for generic image retrieval, incorporating multiple visual features into production search systems delivered to enterprise clients. • Researched hierarchical spatial decomposition approaches for object localization in

About

Detail

Timeline

Résumé