Machine Learning Engineer (Video Foundation Models & Agentic AI)

Description

The company is developing a new generation of model for first-person video, aimed at recognizing what people do, how objects interact, and the physical world.

The goal is video systems that read complex real-world activity accurately and turn continuous visual input into structured information you can act on.

A small startup team covers the full machine learning cycle, from data collection and training through evaluation, optimization, and production deployment, pairing advanced multimodal methods with real-world data and hands-on engineering.

Current systems often fall short on fine-grained activity recognition, temporal reasoning, long videos, and efficiency. The work spans three areas: video foundation models and multimodal understanding, efficient training and deployment, and agentic video understanding with advanced post-training.

Responsibilities

• Design, train, fine-tune, and improve video-language foundation models for practical use.
• Build and extend large-scale ML pipelines for data processing, training, evaluation, and inference.
• Investigate model failures and performance bottlenecks, then improve results through data changes and algorithm work.
• Create inference and deployment approaches that trade off quality, latency, and compute cost.
• Work with product and engineering colleagues to turn business needs into technical plans and production-ready ML features.
• Test new methods in multimodal learning, reinforcement learning, and agentic AI, and apply the useful ones to real problems.
• Own ML projects from problem definition and experiments through deployment and later upkeep.

Requirements

• A bachelor's, master's, or PhD in computer science, artificial intelligence, machine learning, or a related field, or equivalent hands-on experience.
• Solid grasp of deep learning and practical experience training, fine-tuning, or running models.
• Working knowledge of Python and deep learning frameworks such as PyTorch.
• Familiarity with modern LLMs, multimodal models, or computer vision models.
• Strong problem solving, with the ability to build, debug, and optimize ML systems on your own.
• Comfort working at a quick pace, taking ownership, and delivering workable solutions to hard problems.
• Preferred: publications at top venues such as NeurIPS, ICML, ICLR, CVPR, ICCV, or ECCV, or other significant research work.
• Preferred: experience with video understanding, vision-language models, or multimodal foundation models.
• Preferred: experience with LLM or VLM post-training, such as SFT, reinforcement learning, or reasoning-focused training.
• Preferred: experience with distributed training, efficient inference, or production-scale ML systems.
• Preferred: familiarity with agentic AI, tool-augmented reasoning, or multimodal reasoning.
• Preferred: a track record of turning new research ideas into working code and measurable gains.

Benefits

• Real-world impact: your models go into live products and reach users after deployment.
• Full ownership: work across the whole ML lifecycle, from data and modeling to optimization and deployment, and potentially lead meaningful technical initiatives.
• Hard technical problems in video understanding, large-scale training, and efficient inference.
• Close contact with progress in multimodal foundation models, reasoning, reinforcement learning, and agentic AI, with room to try new ideas on real problems.
• A small, fast-moving team that puts ownership, technical quality, rapid iteration, and measurable results first.
• Health insurance
• Vision care
• Dental coverage
• 401(k)
• Paid holidays
• PTO (Paid Time Off)
• Sick leave