machine learning engineer for a proactive AI assistant

Описание

The company is an early-stage AI startup building a proactive assistant for everyday productivity tools such as email, notes, and tasks. Its focus is on reliable long-running workflows, persistent context, and completing tasks despite the non-deterministic nature of underlying models.

Задачи

• Build core machine learning systems powering a proactive, long-horizon AI assistant
• Own ML initiatives end to end, including data preparation, training, evaluation, inference, deployment, and iteration
• Turn research ideas into practical systems that operate reliably in production
• Investigate model failures and system issues using real-world production signals
• Continuously ship, measure results, identify improvements, and refine the system
• Collaborate with research, product, and engineering teams to deliver meaningful user impact
• Mentor other machine learning engineers and provide technical guidance through code reviews and strong engineering practices
• Design solutions accounting for production constraints such as latency, cost, reliability, efficiency, and safety

Требования

• Experience building and shipping machine learning systems used by real users
• Strong understanding of how modern machine learning models behave and fail in production environments
• Excellent Python programming skills and experience with PyTorch and/or JAX
• Experience with GPU-based model training and inference systems
• Ability to write reliable, maintainable, production-quality code
• Strong systems thinking and ability to design solutions beyond experimental scripts or research prototypes
• Ability to work independently, take full ownership, and complete complex initiatives
• Strong problem-solving skills and confidence working with incomplete or ambiguous information
• Clear communication skills
• Fluent English

Условия

• Cash and equity compensation
• Flexible hours in a distributed, global team
• Generous paid time off
• Company laptop provided
• Hiring process includes 3, occasionally 4, interviews, followed by fast decisions