🛠 Tech Stack

💼 About This Role

You'll push the limits of vision-language models for real-world video understanding at an AI-first startup. You'll design and adapt multimodal models and turn them into production pipelines used by customers. This role combines cutting-edge research with applied engineering in a fast-moving team.

🎯 What You'll Do

• Design and adapt vision-language models for video understanding
• Build and maintain large-scale training pipelines on GPU clusters
• Curate and augment video-text and action datasets
• Develop robust benchmarks for video QA and temporal understanding
• Deliver production-ready inference pipelines to product teams

📋 Requirements

• PhD in computer vision, machine learning, or related field
• Strong background in video-centric deep learning
• Experience training large vision or VLM models (e.g., InternVL)
• Proven work with multi-GPU training (PyTorch, distributed)
• Solid engineering habits: clean Python, reproducible experiments

✨ Nice to Have

• Publications at top-tier venues (CVPR, ICCV, NeurIPS)
• Experience with 3D/4D scene representations or action generation
• Inference optimization: quantization, TensorRT, model distillation

🎁 Benefits & Perks

• 💰 Competitive salary & stock options
• 🌍 Collaborative, diverse team with flat hierarchy
• ⏰ Flexible working hours
• 🎯 Real-world impact in manufacturing AI
• 🤝 Supportive culture promoting underrepresented groups