Task Designer, Software Engineering

Reinforced Learning; Town of Tonawanda (NY); On-site; USD 95,000 - 140,000; Full time

Job summary

Reinforced Learning is seeking a seasoned software engineer to build tasks and reference solutions that train models to fix, extend and reason about real codebases. You’ll create real workloads sourced from real repositories, write precise specifications with constraints and acceptance criteria, and review model attempts to identify ambiguities and gaps in solutions.

You’ll collaborate across languages and stacks, bringing strong technical writing and judgement to shape challenging, scalable

Qualifications

• Four or more years as a working software engineer across more than one language and stack.
• Strong technical writing. Your specs are read by people and by models.
• Judgement about difficulty: you know the difference between tedious and hard.
• Interest in how models approach code and where they go wrong.

Skills

Software engineer

Multi-language stacks

Technical writing

Difficulty judgement

Model-aware coding

Job description

The role

A good coding task starts from a real repository and a real failure, not a toy. You’ll write the tasks that teach agents to fix, extend and reason about software the way experienced engineers do, and the reference solutions that prove they can be done.

What you’ll own

• Tasks sourced from real codebases: bugs, migrations, refactors, features with constraints.


• Specifications that state the goal, the constraints and the acceptance criteria without giving away the solution.


• Reference solutions, made checkable with the verifier engineers.


• Task distributions that cover the languages, stacks and difficulty levels a customer actually cares about.


• Review of model attempts to find where tasks are ambiguous, too easy or accidentally leaky.

What you bring

• Four or more years as a working software engineer across more than one language and stack.


• Strong technical writing. Your specs are read by people and by models.


• Judgement about difficulty: you know the difference between tedious and hard.


• Interest in how models approach code and where they go wrong.

Helpful, not required

• Maintainer experience on an open source project.


• Interview problems, courses or coding assessments you’ve written.