Job Overview
The company is seeking a highly skilled and passionate AI Algorithm Engineer to join the cutting-edge research and development team. Unlike standard applied AI roles, this position is deeply focused on the foundational layer of vision models. You will be instrumental in designing, pre-training, scaling, and fine-tuning our next-generation image foundation models from the ground up. You will not just be utilizing existing open-source models; you will be diving into the underlying architectures to create proprietary, state-of-the-art visual generation capabilities. If you are passionate about the complex mathematics of Diffusion models, the underlying architecture of Vision Transformers, and building robust, multi-modal systems that will directly power the core products, you are wanted to the team.
Duties:
- Vision Foundation Model R&D & Iteration: Lead the underlying algorithm research, pre-training, and continuous optimization of cutting-edge image foundation models (e.g., Stable Diffusion, DiT, ViT, SAM, CLIP).
- Business Scenario Customization & Fine-Tuning: Tailor vision models to core business needs (such as AIGC image generation, image editing, and multi-modal understanding). Lead deep supervised fine-tuning (SFT) and parameter-efficient fine-tuning (e.g., LoRA, ControlNet, Adapters).
- Model Architecture Innovation: Explore and improve existing generative models and visual feature extraction network architectures to enhance image generation quality, stability, resolution, and cross-modal alignment.
- Engineering Deployment & Performance Tuning: Collaborate with the engineering team to implement end-to-end deployment of large models. Optimize inference performance and resolve engineering bottlenecks such as high VRAM usage and slow generation speeds (via inference acceleration, quantization, pruning, etc.).
- Cutting-Edge Technology Tracking: Continuously monitor the latest papers and open-source developments in Computer Vision (CV) and multi-modal large models, rapidly reproducing and translating state-of-the-art algorithms into business value.
Job Requirements
- Educational Background: Bachelor's degree or higher in Computer Science, Artificial Intelligence, Applied Mathematics, or a related field.
- [Core Requirement] Deep Expertise in Vision/Image Foundation Models:
- Extensive R&D experience with large vision models, possessing a deep understanding of the underlying architectures and mathematical principles of Diffusion Models, Vision Transformers (ViT), MAE, etc.
- Led or participated as a core member in the full pre-training lifecycle of vision or multi-modal models with hundreds of millions of parameters or more. Highly familiar with data cleaning strategies and ensuring training stability.
- Rich practical experience in Generative AI (AIGC), proficient in the underlying control and optimization technologies for Text-to-Image and Image-to-Image generation.
- Coding & Framework Proficiency: Solid programming foundations, proficient in Python and C++. Highly skilled in the PyTorch deep learning framework with excellent abilities in source code reading and secondary development.
- Distributed Training Experience: Familiar with large-scale cluster parallel training technologies (e.g., DeepSpeed, Megatron, FSDP), with hands-on experience in multi-node/multi-GPU distributed training tuning and compute resource management.
Bonus Points:
- High-quality publications related to large vision models or generative models in top-tier academic conferences (e.g., CVPR, ICCV, ECCV, NeurIPS, ICLR).
- Active contributions to open-source communities (GitHub/Hugging Face) or top-ranking achievements in premier AI algorithms competitions (e.g., Kaggle).
- Experience with low-level operator optimization and High-Performance Computing (HPC) deployment, such as TensorRT and CUDA programming.
Pay: $100,000.00 - $200,000.00 per year
Benefits:
- 401(k)
- Dental insurance
- Flexible schedule
- Health insurance
- Paid time off
- Retirement plan
- Vision insurance
Work Location: Hybrid remote in Arcadia, CA 91007
