Lead AI Research Scientist - Post-Training (Remote)

Startupvalleys; Lausanne; Hybrid; CHF 150,000 - 230,000; Full time

Benefits offered by this job

Remote work option

Job summary

Giotto.ai, a Swiss AI company, seeks a Lead Research Scientist to own post-training, RLHF and policy consolidation from pretrained checkpoints to production models in Lausanne. You will design, implement, scale, and operate the methods for capable, reliable, and controllable production systems while guiding researchers and engineers.

This role is hands-on and strategic, requiring multi-machine training, memory and distribution considerations, and collaboration with data, evaluation, and

Qualifications

• Experience leading large-scale language-model training across multiple machines.
• Strong Python and PyTorch expertise.
• Experience with distributed training and efficient memory management.
• Ability to design end-to-end post-training pipelines.
• Experience with RLHF/RLAIF and reward modelling.

Responsibilities

• Own end-to-end post-training pipeline from pretrained checkpoint to production candidate.
• Set technical direction for post-training and reinforcement-learning work.
• Design multi-stage curricula combining SFT, reinforcement learning, distillation.
• Scale training across multiple machines and accelerators.
• Mentor researchers and engineers and contribute clean code and runbooks.

Skills

Python

PyTorch

Distributed training

Reinforcement learning

Fine-tuning (SFT)

RLHF

Model evaluation

Tools

PyTorch Distributed

FSDP

DeepSpeed

Megatron-Core

Hugging Face Transformers

CUDA

NCCL

Ray

Kubernetes

Docker

MLflow

Weights & Biases

Job description

Giotto.ai, a Swiss AI company, seeks a Lead Research Scientist to own post-training, RLHF and policy consolidation from pretrained checkpoints to production models in Lausanne. You will design, implement, scale, and operate the methods for capable, reliable, and controllable production systems while guiding researchers and engineers.

This role is hands-on and strategic, requiring multi-machine training, memory and distribution considerations, and collaboration with data, evaluation, and