Senior Post-Training AI Scientist (RL & Scaling)

Giotto.ai; Lausanne; Hybrid; CHF 150,000 - 210,000; Full time

Job summary

Giotto.ai in Switzerland seeks a Lead Research Scientist to own and lead the post-training stack, including supervised fine-tuning, RLHF, and verification components. You will design scalable training systems, manage memory and parallelism, and mentor researchers while remaining hands-on.

The role emphasizes end-to-end pipelines, multi-machine training, and collaboration with data, evaluation, and infrastructure teams to deliver reliable, controllable production models.

Qualifications

• Experience leading large-scale language-model training or post-training runs across multiple machines.
• Proven ability to design and run end-to-end post-training pipelines in production.
• Strong Python and PyTorch proficiency with distributed execution expertise.
• Hands-on experience with DeepSpeed, Megatron-Core, or equivalent frameworks.

Responsibilities

• Own the end-to-end post-training pipeline from pretrained checkpoint to production candidate.
• Set technical direction for post-training and reinforcement-learning work.
• Design and execute full-parameter and parameter-efficient SFT.
• Implement preference optimisation, RLHF, and related methods.
• Develop training strategies for reasoning, tool use, multilingual behaviour, and long-horizon tasks.
• Integrate reward models and verifiers; build scalable rollout-generation systems.
• Scale training across multiple machines and accelerators with suitable parallelism.

Skills

LLM training

Python

PyTorch

Distributed training

DeepSpeed

Transformers

CUDA/NCCL

Reinforcement learning

Experiment design

Parallelism strategies

Tools

DeepSpeed

Megatron-Core

Hugging Face Transformers

CUDA

NCCL

Docker

GCP

Weights & Biases

Job description

Giotto.ai in Switzerland seeks a Lead Research Scientist to own and lead the post-training stack, including supervised fine-tuning, RLHF, and verification components. You will design scalable training systems, manage memory and parallelism, and mentor researchers while remaining hands-on.

The role emphasizes end-to-end pipelines, multi-machine training, and collaboration with data, evaluation, and infrastructure teams to deliver reliable, controllable production models.