Requirements

Strong Python engineering skills
Experience carrying a model from raw data through to production
Ability to treat data as the product, focusing on coverage, diversity, and leakage
Design models with awareness that they may exploit weak reward signals
Comfortable working with GPUs and understanding their limitations
Good judgment on when to train models and when not to
Experience working collaboratively in a remote environment

Nice to Haves

Experience with post-training techniques: fine-tuning, reward design, reinforcement learning such as GRPO
Familiarity with RL and fine-tuning frameworks like TRL, verl, or OpenRLHF or custom training loops
Experience with fast rollouts using vLLM or SGLang, and multi-GPU training with FSDP
Experience training tool-using or multi-turn agents
Background in building execution sandboxes, verifiers, evaluation harnesses, or tooling for other engineers
Knowledge of open-weight model families such as Qwen or Llama, and techniques like LoRA

Building environments and verifiers that models train against
Owning the synthetic data pipeline from generation through quality gates
Running end-to-end training experiments and explaining model changes
Building evaluations that releases must pass
Choosing and adapting open-weight base models for specific tasks
Ensuring trained behaviors hold for both voice and text agents
Shipping models to production and continuously improving them based on real usage