← All roles

Member of Technical Staff, Post-Training, RL

Mirendil· Tech
Apply on company site
San Francisco$300k - $400kFull-timePosted 4mo ago

About the role

Research engineer role focused on building the post-training stack for frontier reasoning models, combining research and engineering to turn base models into capable, reliable systems through RL, SFT, and distillation, with emphasis on experimentation, scaling, verification, and end-to-end pipelines.

Responsibilities

  • Develop and iterate on RL, SFT, and distillation recipes; optimize objectives, data mixtures, hyperparameters, rollout generation, and filtering to improve efficiency, stability, and model behavior.
  • Scale post-training workflows to larger scales (tokens, trajectories, models, compute) and identify bottlenecks outside small-run regimes.
  • Train models on long-horizon tasks and develop methods for feedback across long trajectories with sparse rewards, credit assignment, and verification.
  • Work with off-policy and asynchronous training regimes; analyze when off-policy data helps or hurts and manage instabilities.
  • Build robust verification pipelines to detect and reduce reward hacking, false positives, and brittle verifiers.
  • Scale recipes across multiple task families and domains; balance specialization and generality.
  • Conduct experiment analysis and debugging to diagnose regressions, design better ablations, and improve interpretability of post-training results.
  • Collaborate with systems, infrastructure, and data teams to take experiments from idea to production-scale runs; ensure data and verifier quality and reproducibility.

Qualifications

  • Strong engineering skills in addition to research mindset
  • Experience with RL, supervised fine-tuning, and model distillation
  • Proven ability to design and run large-scale experiments and analyze results
  • Ability to work across data, training dynamics, verification, and infrastructure
  • Experience with long-horizon reasoning and off-policy/asynchronous training preferred

Skills

Required:Post-training RLSFT (supervised fine-tuning)Experiment design & analysisVerification & reward quality
Nice to have:DistillationLong-horizon reasoning

Benefits

  • Competitive base salary and meaningful equity grant
  • Competitive benefits
Last updated Aug 27, 2026. Interested? Apply on the company site.

About Mirendil

Mirendil — a frontier lab building systems that excel at AI R&D.

AI R&Dfrontier AIdemocratize frontier AI R&Dsystems that excel at AI R&D
Company profileWebsite

More open roles at Mirendil