Senior Vision-Language AI Scientist for Smart Home Video

Wyze; Kirkland (WA); On-site; USD 134,000 - 181,000; Full time

Job summary

Wyze seeks an Applied Scientist to drive in-house vision-language models for smart home video understanding. You will translate breakthrough research into production VLM development and shape the next generation of physical AI at scale.

Work with tens of millions of authorized videos to uncover event patterns and inform product direction, with opportunities to publish and share open-source models and datasets with the community.

Qualifications

• Hands-on experience training multimodal vision-language models.
• Proficiency with Python and PyTorch.
• Experience with visual transformers or physical world foundation models.
• Strong research sense to identify impactful directions and translate them to practice.

Responsibilities

• Follow the latest breakthroughs in multimodal research and apply them to Wyze VLM development.
• Train, fine-tune, and evaluate vision-language models on large-scale real-world home video data.
• Design rigorous evaluation pipelines for video understanding tasks like event detection and temporal reasoning.
• Analyze user event patterns across millions of videos to guide model design and product direction.
• Contribute to architecture and training of a physical smart home foundation model with advances in embodied AI.
• Build rapid proofs of concept and move promising ideas to validated prototypes.
• Publish research at top venues and contribute open-source models and datasets.
• Help define research problems and set technical direction for industry-relevant solutions.

Skills

Hands-on experience with multimodal V×

Education

PhD in Computer Vision or ML

Job description

Wyze seeks an Applied Scientist to drive in-house vision-language models for smart home video understanding. You will translate breakthrough research into production VLM development and shape the next generation of physical AI at scale.

Work with tens of millions of authorized videos to uncover event patterns and inform product direction, with opportunities to publish and share open-source models and datasets with the community.