Machine Learning Engineer, Driving Foundation Models

Posted: 3 weeks ago
Job Location
Remote Work Policy

In office

Visa Sponsorship

Not Available

RelocationAllowed
Skills
Python
C++
Azure
Docker
AWS
Kubernetes
TensorFlow
Google Cloud
PyTorch
Golang
Jax
DeepSpeed
vLLM
Megatron-LM
Hugging Face Ecosystem

About the job

We are looking for the bestAbout Us42dot is a mobility AI company committed to solving mobility challenges with software and AI. As the Global Software Center of Hyundai Motor Group, 42dot pioneers the future of mobility by advancing the development of software-defined vehicles.We develop safety-first, user-centric software-defined vehicle technologies that deliver the latest performance through continuous updates like smartphones. By advancing software and AI technology, 42dot envisions a world where everything is connected and moves autonomously through a self-managing urban transportation operating system.About The RoleDevelop driving and physical AI foundation models. Support Aria VLA model development.Responsibilities Data Curation & Pretraining: Design and manage massive, multi-terabyte datasets, including data scraping, filtering, deduplication, tokenization, and synthetic data generation.Distributed Model Training: Scale training across massive GPU/TPU clusters using parallel processing, custom data loaders, and framework optimization (PyTorch, JAX, or TensorFlow).Post-Training & Alignment: Implement instruction tuning, Reinforcement Learning from Human Feedback (RLHF), or direct preference optimization to align models with safety and behavior goals.Inference & Optimization: Compress and optimize models (quantization, pruning, speculative decoding) to run at low latency on cloud infrastructure or edge hardware like smartphones and robotics.Evaluation & Guardrails: Build benchmarking harnesses, hallucination checkers, and red-teaming pipelines to test model accuracy and safety.QualificationsLanguages: Python, C++, and Golang (for high-throughput systems).Frameworks: PyTorch, JAX, Hugging Face ecosystem, vLLM.Infrastructure: Kubernetes, Docker, multi-node distributed training (Megatron-LM, DeepSpeed), and cloud platforms (AWS, Azure, Google Cloud).Math & Theory: Linear algebra, calculus, probability, statistics, and deep learning architectures (Transformers, diffusion models).