Research Engineer

Build and productionize post-training systems for voice and text agents, including environments, verifiers, synthetic data, evaluations, and model training. The role requires strong Python and end-to-end model development experience, with reinforcement learning and distributed training expertise preferred.

About the job

Responsibilities

Build the environments and verifiers used to train models.
Own the synthetic data pipeline from generation through quality gates.
Run end-to-end training experiments and explain model changes.
Build release evaluations and define the criteria models must meet.
Choose and adapt open-weight base models for assigned tasks.
Ensure trained behavior works across voice and text agents.
Ship models to production and improve them using real-world usage.

Requirements

Strong Python engineering skills.
Experience carrying a model from raw data through production.
Strong data quality practices, including coverage, diversity, and leakage prevention.
Ability to design against weak or exploitable rewards.
Comfort working with GPUs and understanding their limitations.
Ability to determine when to train and when not to.
Comfortable collaborating in a remote environment.

Nice to Have

Experience with post-training, including fine-tuning, reward design, or reinforcement learning such as GRPO.
Experience with fine-tuning frameworks such as TRL, verl, OpenRLHF, or a custom training loop.
Experience with fast rollouts using vLLM or SGLang.
Experience with multi-GPU training using FSDP.
Experience training tool-using or multi-turn agents.
Experience with execution sandboxes, verifiers, evaluation harnesses, or tooling used by other engineers.
Familiarity with open-weight model families such as Qwen or Llama.
Familiarity with LoRA and similar techniques.