We build multimodal data
and evaluations.

Frontier models are moving beyond text. The next generation of systems will need to reason across vision, sound, time, and tools, but that requires a new kind of data.

Seldon builds multimodal data, computer-use environments and evals for frontier AI-agents. We help models learn across image, video, audio, and text, and test their capabilities in realistic scenarios. Human expertise shapes everything we build, ensuring the quality that is needed to train and evaluate more capable, safe models.