Director of RL & Agentic Post-Training for Production LLMs

Blue Yonder; Paris; On-site; EUR 120,000 - 180,000; Full time

Job summary

Blue Yonder seeks a Director of Reinforcement Learning & Agentic Post-Training to lead how we train LLM-based agents for autonomous supply chain software. This role anchors the Model Training Factory and collaborates with product, engineering, and domain experts to build tool-using agents that operate enterprise software through APIs and workflows.

The candidate will shepherd RL environments, reward modeling, evaluation harnesses, and multi-step workflows across planning and operations.

Qualifications

• Led a team to ship LLM models trained with RL, SFT, DPO, RLHF/RLAIF in production.
• Led teams to train models to use tools, call APIs, or operate multi-step tasks.
• Strong ML engineering background enabling leadership of engineers.

Responsibilities

• Define technical strategy for RL, post-training, and tool-using LLM agents.
• Build and manage ML engineers for agent training, environments, evaluation, data generation, and training infra.
• Design environments where LLM agents interact with Blue Yonder software via APIs and tools.
• Develop training/evaluation systems for multi-step supply chain workflows across planning, WMS, transport, and network ops.
• Define success criteria for operational agents including tool use, constraints, latency, cost, and reliability.
• Create reward models, verifiers, and evaluation harnesses for agent behavior.
• Measure agent performance: tool-call correctness, workflow completion, recovery from bad state, and long-horizon reliability.
• Collaborate with product, architecture, and domain experts to convert real workflows into trainable environments.
• Guide model improvements across supervised fine-tuning, policy optimization, and data generation.
• Balance model capability, inference cost, latency, safety, and product timelines.

Skills

Python

PyTorch

Reinforcement Learning

Leadership

ML Engineering

Tools

NVIDIA Nemotron

NVIDIA NeMo

Megatron

vLLM

Ray

Job description

Blue Yonder seeks a Director of Reinforcement Learning & Agentic Post-Training to lead how we train LLM-based agents for autonomous supply chain software. This role anchors the Model Training Factory and collaborates with product, engineering, and domain experts to build tool-using agents that operate enterprise software through APIs and workflows.

The candidate will shepherd RL environments, reward modeling, evaluation harnesses, and multi-step workflows across planning and operations.