Seeking an ML Engineer for my Online RL team for GigaChat.

We are developing post-training for large language models, including various online RL approaches, and are looking for someone who can not only implement existing ideas but also independently lead work from hypothesis to measurable model quality improvement.

What you will do:

• develop and improve online RL methods: RLHF and GRPO family algorithms
• build a complete training loop: rollouts, reward signals, verifiers, and model updates
• design experiments and understand why a result worked or didn't work
• develop reliable and reproducible training infrastructure
• optimize distributed training, throughput, and GPU utilization

We are looking for a strong engineer with good knowledge of Python and PyTorch, experience in LLM post-training / online RL / RLHF, and the ability to independently conduct ML experiments from hypothesis to conclusions.

Will be a plus: experience with GRPO and similar policy optimization approaches, reward models, LLM-as-a-judge, verifiers for general-domain tasks, as well as with vLLM, SGLang, verl, Megatron, or TRL.

Why join us:

• direct impact on GigaChat quality
• combination of research and engineering work
• complex tasks on large models and scalable infrastructure
• strong team of engineers and researchers

If you are interested in developing online RL and turning research ideas into working solutions, write to me in direct messages.