Machine Learning Engineer
— Pinterest
Toronto, ON | Jan 2024 - Present
- Developing cutting-edge GenAI pipelines for visual discovery and content recommendation, utilizing latent diffusion models to generate context-aware backgrounds for Product Pins.
- Reduced inference cost by 83% (saving ~$2M annually) via FP8/FP16 mixed-precision quantization for the online serving pipeline.
- Implemented dynamic LoRA refitting within TensorRT engines, enabling multi-style image generation for millions of users without model reloading.
- Trained GAN-based enhancement models to achieve sub-pixel precision in image retrieval, significantly improving recommendation relevance.