Machine Learning Engineer
Current- Executed end-to-end deployment of multiple generative AI text-to-image and text-to-video models, including SD 1.5, SD 2.1, SDXL, and SDXL Lightning, for Imagine.ai, Animagic.ai, Reshot, and PhotoTune, boosting product suite downloads to over 25 million globally.- Optimized Performance: Reduced inference time & server costs by 50%, enhancing the processing of over 70 million images, handling 17 million requests every day, managing 16 TB of data while serving 10 million monthly active users.- CI/CD Implementation: Designed CI pipelines for feature integration testing, achieving a 95% reduction in production errors.- Cluster Deployment Expertise: Demonstrated hands-on expertise in cluster deployments, from architecture to management.- Carried out deployment of Animagine XL and multiple anime-style high-resolution text-to-image models based on Stable Diffusion XL, for the Animagic with over 1 million downloads. Fine-tuned with a learning rate of 4e-7 over 27,000 steps on high-quality anime-style images.- Carried out the optimization and migration of all ComfyUI-based features, including Face Avatar, Inpaint, Headshot, Enhance, BGRemover, AIFilters, Inpaint Remover, various text-to-image and image-to-image Stable Diffusion features into Python, achieving an optimized inference speed improvement of up to 40%.- Proudly open-sourced fine-tuned text-to-image model (v4) on Hugging Face Spaces, leveraging Gradio and AWS SageMaker to deliver cutting-edge AI capabilities to the community.- Developed and deployed the headshot feature using BrushNet and MediaPipe, engineered an advanced faceavatar workflow leveraging IPADAPTER, SDXL models, latent interposer, and ksampler for enhanced facial representation and customization, and deployed an inpaint feature utilizing Stable Diffusion v1, a latent diffusion model combining an autoencoder with a diffusion model trained in the latent space of the autoencoder, all integrated seamlessly into a ComfyUI workflow.
