Member of Technical Staff, Model Training (post-training research)

Role: Model Training Member of Technical Staff (post-training research)

Type of job: full-time

Location: Remote

About the role

You'll study how training data, evaluation methods, and learning algorithms combine to give models better reasoning, generalization, and agent skills. You'll try new approaches to RL, distillation, and data-efficient training, and turn research findings into methods that scale and make models smarter.

The job is highly experimental, sitting between core ML research and large-scale model building. You'll form research hypotheses, design ambitious experiments, and create training methods that shape what micro1's models can do.

What You'll Do

• Study, code, and scale up advanced post-training approaches (SFT or supervised fine-tuning, RL or reinforcement learning, and OPD or on-policy distillation) for big language and multimodal models.
• Study credit assignment, exploration strategies, optimization dynamics, and learning objectives to make training more stable and sample-efficient and help models gain capabilities.
• Build and refine training recipes by tuning data mixes, sampling strategies, curricula, and evaluation methods together.
• Design ways to learn from human preferences, expert demonstrations, corrections, and trajectories produced by models, and find where human feedback helps most.
• Study how generators, reward models, and graders evolve together, and develop methods that lift model performance as well as the quality of the training signals models learn from.
• Design careful experiments, ablations, and eval frameworks that measure reasoning, generalization, and performance on realistic tasks.
• Own research projects end to end, from picking promising directions and coding new algorithms to running large experiments and confirming results.

What We're Looking For

• Strong grounding in deep learning, machine learning, optimization, and current training methods.
• Practical experience post-training, fine-tuning, or training large multimodal or language models.
• Deep expertise in at least one area such as distillation, RL, data selection, learning dynamics, or reward modeling.
• Strong Python and PyTorch, able to build, debug, and tune complicated training pipelines.
• Background designing controlled experiments, reading model behavior, and turning findings into research directions to act on.
• Can work through unclear research problems alone, form hypotheses, and iterate fast on the evidence.
• Strong analytical, research-minded thinking that values reproducibility, scientific rigor, and measurable gains.

Nice to Have

• Background with RLHF, RLAIF, preference optimization, or RL for reasoning and agent tasks.
• Knowledge of distributed training, large experiments, and GPU-based model tuning.
• Background building graders, verifiers, reward models, or automated evaluation.
• Research contributions in post-training, RL, or nearby areas through papers, open-source work, or production systems.
• Background studying curriculum learning, iterative self-improvement methods, or synthetic data generation.

Compensation & Benefits Notice

Nationally, base pay for this full-time job runs $200,000–$320,000. Every employee can receive equity, and bonuses tied to performance may also apply depending on the role and company policy. Benefits cover health-insurance premiums (reimbursed at up to 100%), paid time off, a 401(k) with a company match, and other perks meant for a remote-first team that performs at a high level.

micro1 hires on an equal-opportunity basis. Every qualified applicant is considered without regard to age, color, disability, national origin, race, religion, sex (pregnancy, gender identity, and sexual orientation included), veteran status, genetic information, or any other trait that applicable local laws, ordinances, or regulations protect. Applicants who need help or a reasonable accommodation while applying can ask for it.

AI tools help screen and assess candidates in the hiring process. They are meant to support people's hiring decisions, not to take them over.

Application process

Submit your application through NorthStar. Employer screening may follow after submission.