🛠 Tech Stack

💼 About This Role

You'll own the layer that makes petabytes of video queryable by content, running vision-language models over every clip to enable researchers to find relevant data in minutes. Your work directly accelerates customer model training iterations. This role combines research and engineering in a small, tight-knit team backed by top investors and partnered with leading Physical AI labs.

🎯 What You'll Do

• Own the visual understanding roadmap end-to-end.
• Train, fine-tune, and evaluate VLMs and embedding models.
• Drive down per-clip annotation cost at corpus scale.
• Design taxonomies and instrument quality for customer datasets.

📋 Requirements

• Strong familiarity with modern vision and multimodal models (VLMs, VQA, embeddings).
• Experience running these models at scale on real video/sensor data.
• Background from a perception team at a self-driving, robotics, or visual-data company.
• Comfortable with cloud infrastructure and large-scale data processing.

✨ Nice to Have

• Experience training vision or multimodal models from scratch.
• Hands-on time with big-data frameworks like Spark, Ray, or Daft.
• Experience designing labeling taxonomies or running annotation programs.

🎁 Benefits & Perks

• 🏥 Health, vision, and dental coverage
• 🏖️ Flexible PTO
• 🍱 Catered lunches and dinners
• 🚆 Commuter benefit
• 💻 Latest Apple equipment

📨 Hiring Process

Estimated timeline: 2-4 weeks · AI estimate

1. 1Recruiter Screen· 30 min
2. 2Technical Interview· 60 min
3. 3On-site Loop· 3 hours