AI/ML Research Engineer

Hanzo AI; United States; Remote; USD 180,000 - 240,000; Full time

Job summary

Hanzo AI is seeking a scientist-engineer to train and improve foundational AI models across pretraining, fine-tuning, evaluation, and deployment. The role blends research with systems engineering, touching text, code, multimodal, and agent-oriented applications.

You will collaborate with teams to run distributed GPU training, build robust evaluations, and deliver reusable outputs such as model weights and reproducible recipes.

Qualifications

• Strong understanding of transformer architectures and training methods.
• Experience with PyTorch and distributed training at scale.
• Proven production ML delivery or open-source work.
• Systems knowledge: concurrency, networking, memory efficiency.

Responsibilities

• Train and fine-tune foundation models with human or AI feedback.
• Plan and run distributed GPU training across clusters.
• Build evaluations for reasoning, tool use, and safety.
• Collaborate with inference and compiler teams to deploy models.
• Produce reusable outputs such as weights, papers, and reproducible training recipes.

Skills

Transformers

PyTorch

Distributed training

GPU profiling

Open-source

Research papers

Systems design

Model evaluation

Tools

Triton

CUDA

CUDA kernels

Job description

Role overview

Train and improve foundational AI models, from pretraining and fine-tuning to evaluation and production deployment. The role spans research and systems engineering, with work across text, code, multimodal, and agent-oriented applications.

Responsibilities

• Train and fine-tune foundation models, including workflows involving human or AI feedback.
• Plan and run distributed training across GPU clusters.
• Build evaluations for reasoning, tool use, and safety.
• Work with inference and compiler teams to optimize and deploy models.
• Produce reusable research outputs such as model weights, papers, and reproducible training recipes.

Requirements

• Strong understanding of transformer architectures, attention, and current language-model training methods.
• Practical experience with PyTorch, distributed training frameworks, and GPU performance profiling.
• Evidence of delivering production machine-learning models, substantial open-source work, or published research.
• Systems knowledge covering concurrency, networking, and memory efficiency.

Nice to have

• Experience with synthetic data, agent reasoning methods, or tool-use tuning.
• Familiarity with Triton, CUDA kernel development, or model quantization formats.

Benefits and work setup

• Remote-first work across distributed locations.
• Emphasis on measurable outcomes, open-source contributions, and individual ownership.