At Scale AI, we label on the order of 10MM annotations per week. Our data is diverse both in image space (e.g. cameras, weather conditions, driving surface) and label space (e.g. object and attributes categories). To fully leverage the vast amount of labeled data, we want to continuously fine-tune a base model across all datasets as tasks are completed. The base model allows us to easily fine-tune it further for downstream tasks such as quality assurance, prelabeling, and image similarity search. We present a multi-task continuous learning approach that can be trained efficiently as more labeled data becomes available. In conjunction to this, we also provide real-time annotation solutions with our models as a service offering for document intelligence. We utilize our labeling platform, our machine learning expertise, and our customer’s domain knowledge to provide a customer-specific, problem-specific, real-time data labeling solution.