Research Engineer, Post-Training

Reinforced Learning; Town of Tonawanda (NY); On-site; USD 120,000 - 160,000; Full time

Job summary

Reinforced Learning seeks ML engineers to build end-to-end RL/LM training pipelines in a research engineering setting. You’ll implement SFT and RL post-training workflows to isolate dataset effects and run comparable evaluations.

You’ll design experiment tracking, checkpoint management, and evaluation plumbing, and integrate with customer training stacks where programs require it, delivering concise reports with confounders named.

Qualifications

• Two or more years in ML engineering or research engineering.
• Hands-on fine-tuning of language models and running evaluations.
• Strong Python and systems debugging across data, compute and model behaviour.
• You ship reproducible pipelines.

Responsibilities

• SFT and RL post-training pipelines on open models.
• Controlled experiments that measure what a dataset changes.
• Experiment tracking, checkpoint management and evaluation plumbing.
• Integration with customer training stacks where programs require it.
• Concise reports with the confounders named.

Skills

ML engineering

Python

System debugging

Reproducible pipelines

Fine-tuning language models

Tools

Distributed training frameworks

Cluster schedulers

Job description

The role

Our claim is that our data improves models. You’ll build the infrastructure to prove it: fine-tuning and RL runs that isolate the effect of a dataset, evaluation that is comparable across runs, and tooling that makes the next experiment faster than the last.

What you’ll own

• SFT and RL post-training pipelines on open models.
• Controlled experiments that measure what a dataset changes.
• Experiment tracking, checkpoint management and evaluation plumbing.
• Integration with customer training stacks where programs require it.
• Concise reports with the confounders named.

What you bring

• Two or more years in ML engineering or research engineering.
• Hands-on fine-tuning of language models and running evaluations.
• Strong Python and systems debugging across data, compute and model behaviour.
• You ship reproducible pipelines.

Helpful, not required

• Distributed training frameworks and cluster schedulers.
• RL algorithms for language models.