AI Engineer, RL Post-Training and Agentic Systems

In sede · Munich (Germania) · Mid · Tempo pieno

Vuoi altre offerte come questa?

Crea il tuo profilo gratuito su PyHires: vedrai quanto sei compatibile con ogni offerta e riceverai ogni settimana un'email con quelle più adatte a te.

Crea il mio profilo

Panoramica

Modalità di lavoro
In sede
Sede
Munich (Germania)
Tipo di contratto
Tempo pieno
Seniority
Mid
Lingue
Inglese
Aperta a candidati da
Solo Paesi UE
Azienda
Vesoma
Pubblicata
19 nov 2026

Il ruolo

Vesoma is building humanoid robots for physical labour, engineered to automotive standards, with its two main locations in Munich and Limassol. You will join the early AI team to build the infrastructure that post-trains its robot foundation models with large-scale reinforcement learning and deploys them as agentic systems across the robot fleet, owning the loop from distributed rollouts to serving models on cloud and on-prem infrastructure. This is a full-time, permanent position. What you'll do - Build scalable infrastructure for RL post-training of multi-billion-parameter models, coordinating actors, learners and reward models across GPU clusters, simulation and robots - Own cloud and on-prem model serving with low-latency, high-throughput inference for RL rollouts and production robots, including fast weight synchronisation between training and inference - Build the infrastructure to orchestrate, evaluate and monitor agentic systems that combine models with planning, tool use and robot skills What you bring - 3+ years building and operating large-scale RL or post-training systems for large models (actor-learner architectures, asynchronous rollouts, off-policy correction) - PPO, GRPO and related algorithms, reward modelling and frameworks such as veRL, OpenRLHF, NeMo-RL, TRL or Ray RLlib - Inference engines such as vLLM, SGLang or TensorRT-LLM, batching, KV-cache optimisation and quantisation - Ray, Kubernetes and/or Slurm, with monitoring in Prometheus and Grafana - Python, PyTorch and/or JAX, CUDA, NCCL and distributed training with FSDP or Megatron-Core Nice to have - Agentic systems and evaluation harnesses for long-horizon, multi-step tasks - Robot learning in simulation and sim-to-real transfer (Isaac Sim/Isaac Lab, MuJoCo) - Edge inference on NVIDIA Jetson/Orin/Thor, TensorRT or ONNX Runtime - ROS 2 and MCAP

Requisiti

Basata su:Python

Indispensabili

  • Grafana
  • JAX
  • Kubernetes
  • Prometheus
  • PyTorch

Chi è Vesoma

Vesoma is building humanoid robots for physical labour, engineered to automotive standards of safety, scale and durability, with the goal of becoming a European leader in physical AI. It is well capitalised and has two main locations, in Munich and Limassol.

Stack

  • Python
  • Grafana
  • JAX
  • Kubernetes
  • Prometheus
  • PyTorch

Azienda