Key Responsibilities

• Design and implement large-scale multi-modal architectures (e.g., vision–language–action transformers) for end-to-end autonomous driving.
• Develop pretraining and fine-tuning strategies leveraging massive labeled and unlabeled fleet data (images, video, LiDAR, CAN bus, maps, human driving behaviors, etc.).
• Research and integrate cross-modal alignment (e.g., visual grounding, temporal reasoning, policy distillation, imitation and reinforcement learning) to improve model interpretability and action quality.
• Collaborate with infrastructure engineers to scale training across thousands of GPUs using distributed training frameworks (FSDP, DDP, etc.).
• Conduct systematic ablation, evaluation, and visualization of model behavior across perception, reasoning, and planning tasks.
• Contribute to model deployment optimization, including quantization, export, and latency–accuracy trade-offs for onboard execution.

Minimum Qualifications

• Master’s degree or higher in Computer Science, Electrical/Computer Engineering, or related field, with 3+ years of experience in deep learning research or productization.
• Strong proficiency in PyTorch and modern transformer-based model design.
• Experience in large-scale pretraining or multi-modal modeling (vision, language, or planning).
• Deep understanding of representation learning, temporal modeling, and self-supervised or reinforcement learning techniques.
• Familiarity with distributed training (DDP, FSDP) and large-batch optimization.

Preferred Qualifications

• PhD in CS/CE/EE or related field, with 1+ years of relevant industry experience.
• Publication record in top-tier AI conferences (CVPR, ICCV, NeurIPS, ICLR, ICML, ECCV).
• Prior experience building foundation or end-to-end driving models, or LLM/VLM architectures (e.g., ViT, Flamingo, BEVFormer, RT-2, or GRPO-style policies).
• Familiarity with RLHF/DPO/GRPO, trajectory prediction, or policy learning for control tasks.
• Proven ability to collaborate cross-functionally with infra, perception, and planning teams to deliver production-ready models.