Production-worthy computer vision models need large quantities of high-quality training data, even when the models themselves are tiny. Plumerai’s fully in-house data tooling therefore leverages several powerful machine learning techniques, allowing us to build and curate datasets with millions of images. In this talk, Jelmer will demonstrate how these tools allow us to identify and address issues in our dataset that negatively affect our models, for example by inspecting images that contributed strongly to a specific false prediction during training. He will also cover the other essential parts of our model and data pipelines, which together allow us to build accurate person detection models that run on the tiniest edge devices.