Job Description

About the Team

We are building Max VLM, a next-generation video foundation model designed to understand human activities, object interactions, and the physical world through first-person video.

Our mission is to develop intelligent video understanding systems that can accurately interpret complex real-world activities and turn continuous visual observations into structured, actionable information.

As a fast-moving startup team, we work across the entire machine learning lifecycle, from data collection and model training to evaluation, optimization, and production deployment. We combine state-of-the-art multimodal AI technologies with real-world data and practical engineering to build scalable, reliable, and efficient AI systems.

Project Overview & Technical Challenges

Despite rapid advances in multimodal large language models, understanding real-world video remains a challenging problem. Existing systems often struggle with fine-grained activity recognition, temporal reasoning, long-video comprehension, and computational efficiency.

Our team focuses on building high-performance video foundation models and bringing them into real-world applications. Key technical directions include:

1 - Video Foundation Models & Multimodal Understanding

Develop and improve video-language models capable of understanding complex human activities and physical-world interactions.

Fine-grained action recognition, temporal grounding, and video understanding.

Long-video comprehension, temporal reasoning, and event-level understanding.

Multimodal model fine-tuning, post-training, and performance optimization.

High-quality training data generation, automated labeling, and data selection.

Model evaluation, error analysis, and data-driven iteration to continuously improve production performance.

2 - Efficient Model Training & Deployment

Build scalable and efficient machine learning systems that support real-world product requirements.

Large-scale model training, fine-tuning, and distributed training optimization.

Efficient video processing, and inference acceleration.

Model compression, quantization, and optimization of latency, throughput, and computational cost.

Production model serving, monitoring, and continuous performance improvement.

End-to-end optimization of model quality, reliability, and deployment efficiency.

3 - Agentic Video Understanding & Advanced Post-Training

Explore advanced AI techniques to improve video understanding and reasoning capabilities.

Develop agentic video understanding systems with autonomous tool use and multi-step reasoning.

Explore reinforcement learning and advanced post-training techniques to improve model performance.

Apply emerging AI research to real-world applications and product improvements.

Key Responsibilities

Design, train, fine-tune, and optimize video-language foundation models for real-world applications.

Build and improve large-scale ML pipelines covering data processing, training, evaluation, and inference.

Analyze model failures, identify performance bottlenecks, and drive continuous improvements through data and algorithmic iteration.

Develop efficient model inference and deployment solutions that balance quality, latency, and computational cost.

Collaborate closely with product and engineering teams to translate business requirements into technical solutions and deliver production-ready ML capabilities.

Explore and evaluate emerging techniques in multimodal learning, reinforcement learning, and agentic AI, applying promising advances to practical problems.

Take ownership of ML projects from problem definition and experimentation through deployment and ongoing improvement.

Qualifications

Minimum Qualifications

Bachelor's, Master's, or PhD degree in Computer Science, Artificial Intelligence, Machine Learning, or a related field, or equivalent practical experience.

Strong understanding of deep learning and hands-on experience with model training, fine-tuning, or inference.

Proficiency in Python and deep learning frameworks such as PyTorch.