Belvedir
Member of Technical Staff, Training
San Francisco, CA · Staff · Full-time · Posted Oct 29, 2026
- Visa sponsorship
- Visa sponsorship
About the job
Designs and runs customer-specific model-training experiments, builds data and evaluation pipelines, and turns successful training work into reusable recipes and tools.
What you'll do
- Run SFT, distillation, preference-tuning, and RL experiments on customer tasks.
- Build pipelines for collecting, cleaning, filtering, and synthesizing training data, plus task-specific evaluations.
- Work with customers and founders to choose training priorities and develop repeatable training recipes and tooling.
What they ask for
- Experience with LLM training, including post-training, RL, RLHF, evaluations, or synthetic data, in industry or research.
- Experience training or fine-tuning models on real, messy data.
- Knowledge of trustworthy evaluations and how models can exploit flawed evaluations.
- Ability to work directly with customers and translate their needs into training priorities.
Nice to have
- Founder or early-engineer experience.
Tools
LLM, SFT, RL, RLHF
AI focus
Model training and post-training, reinforcement learning, synthetic data, evaluation, and training-data pipeline development.
Model-written summary · the original posting is authoritative