Education & Background

• MSc or PhD in Computer Science, Artificial Intelligence, Mathematics, or a related quantitative discipline. PhD strongly preferred.

• Candidates with publications at top-tier venues (CVPR, ICCV, NeurIPS, SIGGRAPH, etc.) will be prioritized.

• Graduates from leading universities with a strong research component are highly desirable.

Role Summary

Train and optimize large generative models for multimodal-to-3D generation. You’ll lead experiments, drive quality improvements, and deliver robust, production-grade generation pipelines.

Responsibilities

• Train, fine-tune, and align large generative models (image/multimodal-to-3D).

• Manage the full experiment lifecycle: data mixtures, checkpointing, evaluation, and model comparison.

• Improve generation quality in geometry, texture, structure, and semantic alignment.

• Optimize training and inference for GPU clusters; diagnose instability, overfitting, and train-inference gaps.

• Collaborate with CV, 3D, and data teams to define and execute model release workflows.

Requirements

• Deep expertise in deep learning: Transformers, diffusion models, VAEs, conditioning mechanisms.

• Strong PyTorch skills; experience building scalable training pipelines.

• Proficient in distributed training (DDP, FSDP, DeepSpeed, or similar).

• Hands-on experiment tracking, ablation design, benchmark evaluation, and failure analysis.

• Disciplined engineering: reproducible training, config management, logging, model versioning.

Nice to Have

• Experience with 3D generative models, mesh/multi-view generation, or 3D reconstruction.

• Knowledge of low-poly asset generation, mesh extraction, texture baking, UVs, or GLB export.

• Multimodal training, image encoder conditioning, captioning, data filtering, or model fine-tuning.

• Managing large-scale GPU jobs and solving performance/stability bottlenecks.