Data is a fundamental layer in Luma that unlocks advanced capabilities in our foundation models. We tackle the fundamental data questions around how different modalities can be combined to enable new behaviors and capabilities, working on the open-ended challenges of what makes multimodal AI systems truly powerful and versatile.

• Identify capability gaps and research solutions

• Design datasets and data-mixture ablations to systematically improve model capabilities across vision, audio, and language

• Develop evaluation frameworks and benchmarking approaches for multimodal AI capabilities

• Create prototypes and demonstrations that showcase new multimodal capabilities

• Strong programming skills in Python and PyTorch

• Experience with large-scale dataset

• Experience with multimodal data processing pipeline

• Understanding of computer vision, audio processing, and / or natural language processing techniques

• (Preferred) Expertise working with interleaved multimodal data

• (Preferred) Hands-on experience with Vision Language Models, Audio Language Models, or generative video models