Andreas Dengel

Forget expensive, privacy-violating manual annotation: DocDjinn lets you bootstrap high-quality document understanding models from just 100 real samples using VLM-powered synthetic data generation.

Stunning visuals from the latest text-to-image models don't guarantee better training data -- in fact, they can make your image classifiers *worse*.

Routine carotid ultrasound, when analyzed with deep learning, unlocks a wealth of untapped prognostic information, potentially surpassing conventional risk models for cardiovascular disease.