CareerPlan

Researcher - Post-Training, Reinforcement Learning & Training Systems

Lausanne, VaudRemoteFull-time2026-10-07 → 2026-11-16
Hybrid

Core

Design, implement, and operate large-scale post-training pipelines (SFT, RLHF, RL with verifiable rewards) for production AI models, scaling across multiple machines and accelerators.

Role type

Senior Research Engineer / Research Scientist (Post-Training & Training Systems)

Builds

Scalable, reliable, and controllable production AI models via distributed training and optimization.

Domain

Artificial Intelligence / Machine Learning / Distributed Systems

Deliverable

production ML models

Required skills

Large-scale distributed training, Reinforcement Learning (RLHF, RLAIF), Supervised Fine-Tuning (SFT), Reward modeling, Distributed execution strategies, Debugging numerical instability, Experimental design, Python, PyTorch, DeepSpeed, FSDP, Megatron-Core, CUDA, NCCL, vLLM, Ray, Kubernetes, MLflow, W&B

Preferred skills

CUDA or Triton, Long-context training, Sparse models, Asynchronous RL, Stateful agent environments, Policy distillation, Deployment-aware post-training

Technologies

PyTorch, DeepSpeed, FSDP, Megatron-Core, Hugging Face Transformers, CUDA, NCCL, vLLM, SGLang, Ray, Slurm, Kubernetes, MLflow, Weights & Biases, Docker, GCP, GitLab CI, pytest

Responsibilities

Own end-to-end post-training pipelines from checkpoint to production; Design and execute full-parameter and parameter-efficient SFT; Implement preference optimization and reinforcement learning with verifiable rewards; Scale training across multiple machines using parallelism strategies; Build rollout-generation systems for iterative training; Diagnose training failures (OOM, convergence regressions, reward hacking); Build reliable checkpointing, recovery, and monitoring procedures.

Seniority

Senior, hands-on IC