AI Research Engineer

Fuel Talent; Seattle (WA); On-site; USD 147,000 - 220,000; Full time

Benefits offered by this job

Visa support

Open-source collaboration

Seattle-based opportunity

Job summary

Fuel Talent partners with a Seattle-based AI research institute to recruit researchers for training flagship open models. You will own delivery end-to-end from system design to experiment release, collaborating with research and engineering peers.

Requires 4+ years in ML infrastructure, strong Python and PyTorch skills (JAX a plus), and a PhD is a bonus. Opportunity to work on vision-language models, MoE training, and RL-based methods in a cutting-edge environment.

Qualifications

• BS/MS or higher in CS, Math, or a related quantitative field
• 4+ years in ML infrastructure and experience training LLMs or multimodal models end-to-end at scale
• Proficiency in Python & PyTorch (JAX a plus)

Responsibilities

• Build and optimize infrastructure for LLMs, multimodal, and agentic research: training/inference pipelines, dataset curation, large-scale preprocessing
• Design, train, and evaluate multimodal models and agentic workflows, including tool use, planning, and long-horizon tasks
• Scope and lead research projects, prioritizing experiments for highest impact
• Contribute to the open-source community via model releases, datasets, APIs, and technical reports

Skills

ML infrastructure

Python

PyTorch

Large language models

Multimodal models

Research collaboration

Education

BS/MS in CS/Math or related

Tools

GCP

Docker

CUDA

Job description

Compensation: $147-220k base + 5-15% bonus + long-term incentives

Visa Support: yes for exceptional talent

We are partnered with a Seattle-based AI research institute building fully open AI: large-scale models, datasets, and public artifacts across language, multimodal (vision-language), and agentic systems. With academic freedom and corporate-scale compute, our client offers a rare chance to train frontier open models and share the results with the broader research community.

We are looking for Research Engineers to help train the client's flagship open models. From system design through experiment release, you will own delivery end to end while collaborating closely with research and engineering peers.

Tech stack: Python, PyTorch, JAX, GCP, Docker, CUDA, vLLM, SGLang, RL training frameworks, vision-language models, MoEs, post-training (instruction tuning, RL, reasoning)

What you'll do

• Build and optimize infrastructure for LLM, multimodal, and agentic research: training and inference pipelines, dataset curation, large-scale preprocessing
• Design, train, and evaluate multimodal (vision + language) models and agentic workflows, including tool use, planning, and long-horizon tasks
• Scope and lead research projects, prioritizing experiments for highest impact
• Contribute to the open-source community through model releases, datasets, public APIs, and technical reports

Must-Haves

• BS/MS or higher in CS, Math, or a related quantitative field (must-have)
• 4+ years in ML infrastructure and experience training LLMs or multimodal models end-to-end at scale
• Python & PyTorch proficiency (JAX a plus)

Depth in one of the following:

• Mixture of Experts (MoE) training, pretraining (language + multimodal)
• Long-sequence models
• Supervised Fine-Tuning (SFT) dataset building
• Reinforcement Learning (RLVR, GRPO, PPO), RL and agent environments
• Reasoning and agentic model training
• CUDA and compute infrastructure optimization

Bonus Points

• PhD in ML or equivalent deep learning research experience
• Agentic systems or vision-language model experience
• Cloud infrastructure (GCP/AWS, Docker, distributed training)
• Alignment with open-source AI and a non-profit mission