AI/ML Research Engineer

Hanzo; United States; Remote; USD 170,000 - 240,000; Full time

Job summary

Hanzo is seeking an ML research engineer to advance the model layer of an AI platform, training and fine-tuning foundation models, exploring feedback-based learning, and developing reasoning methods across research and production.

You will work on distributed training, evaluation, optimization, and deployment, delivering reproducible outputs such as research, model weights, or training recipes, with remote-first work and high autonomy.

Qualifications

• Deep understanding of transformer architectures, attention, and pre-training/post-training techniques.
• Hands-on experience with PyTorch and distributed training frameworks such as FSDP, DeepSpeed, or Ray.
• Experience profiling GPU performance.
• Evidence of shipping ML models to production or producing strong open‑source or published research.
• Solid systems knowledge, including concurrency, networking, and memory efficiency.

Responsibilities

• Train and fine-tune models for text, code, multimodal, and agent-based tasks.
• Plan and run distributed training jobs across GPU clusters.
• Build evaluation tools and benchmarks for reasoning, tool use, and safety.
• Work with inference and compiler teams to quantize, optimize, and deploy models.
• Contribute research papers, open model weights, and reproducible training materials.

Skills

Transformer architectures

PyTorch

Distributed training

Evaluation benchmarks

Open research contributions

Tools

PyTorch

DeepSpeed

Ray

Triton

Job description

Role overview

Advance the model layer of an AI platform by training and fine-tuning foundation models, exploring feedback-based learning, and developing reasoning methods. The role spans research engineering and production work, from distributed training and evaluation to model optimization and deployment. Work is expected to include reproducible outputs such as research, model weights, or training recipes.

Responsibilities

• Train and fine-tune models for text, code, multimodal, and agent-based tasks.
• Plan and run distributed training jobs across GPU clusters.
• Build evaluation tools and benchmarks for reasoning, tool use, and safety.
• Work with inference and compiler teams to quantize, optimize, and deploy models.
• Contribute research papers, open model weights, and reproducible training materials.

Requirements

• Deep understanding of transformer architectures, attention, and modern pre-training and post-training techniques.
• Hands‑on experience with PyTorch and distributed training frameworks such as FSDP, DeepSpeed, or Ray.
• Experience profiling GPU performance.
• Evidence of shipping machine learning models to production or producing strong open‑source or published research.
• Solid systems knowledge, including concurrency, networking, and memory efficiency.

Nice to have

• Experience generating synthetic data, tuning tool use, or exploring agent reasoning methods such as MCTS or ReAct.
• Familiarity with Triton, CUDA kernel development, or custom model quantization formats.

Benefits and work setup

• Remote‑first work across a distributed team, with outcomes prioritized over time at a desk.
• High autonomy and ownership, with opportunities to contribute work for public inspection and reuse.