Founding Engineer, Model Training and Research

Unbounded needs the model-training capability to turn human work traces into useful training data and improve the reliability of agents performing complex, long-horizon tasks. This hire bridges research, training infrastructure, evaluation, and practical deployment.

You will design and run open-model pre-training and post-training experiments, develop representations of workflows and decisions, and build evaluations for long-running agent work. The single most important spike is hands-on model training and evaluation that produces meaningful, reproducible improvements for agent performance.

What You'll Own

• Design and run experiments using open models, applying pre-training or post-training approaches as appropriate.
• Build systems that turn human work traces into useful training data.
• Explore representations of workflows, context, and decisions that improve agent performance.
• Develop evaluations that measure quality and reliability across long tasks.
• Build reproducible experimentation and training workflows.
• Partner with infrastructure engineers to bring successful approaches into deployed systems.
• Own research questions from hypothesis through experimentation and practical implementation.
• Diagnose training failures and distinguish meaningful improvements from misleading results.

Technical Requirements

• Hands-on experience with open models and depth in pre-training, post-training, or both.
• Strong programming skills and ability to implement and evaluate research ideas.
• Understanding of training data, optimization, evaluation, and experimental design.
• Ability to build the engineering systems needed to support research.

Execution and Ownership

• Own research questions end to end.
• Choose approaches based on evidence and the problem being solved.
• Work directly with the founder and collaborate closely with infrastructure engineers.
• Operate effectively with uncertain outcomes and high autonomy.

Background

• Around 2–3 years of relevant work is the target
• Experience at an applied AI, open-model, or post-training company is relevant but not required.
• Strong independent research and demonstrated technical work count; academic credentials are flexible.

Location and Visa

• Work in person five days per week.
• New York is preferred; San Francisco or Toronto may be considered.
• Candidates elsewhere must be willing to relocate; permanent unwillingness to relocate is not aligned with the first-hire expectation.
• Visa sponsorship and permitted statuses: — pending / confirm.

Nice to Have

• Experience training models for agent tasks or learning from behavior and workflow traces.
• Experience with training efficiency or distributed training infrastructure.
• Public experiments, open-source contributions, or research implementations.
• Startup or entrepreneurial experience.

Who Will Thrive Here You enjoy connecting fundamental questions to practical experiments, are rigorous about evaluating results, and are willing to own the engineering required to make a promising idea useful. The role suits a scrappy, entrepreneurial builder who wants substantial autonomy and responsibility at the beginning of a company.