Software Engineer – Terminal Bench (Remote) | Immediate Start

About Dusker AI

Dusker AI is an innovation-driven company advancing the reliability and real-world performance of AI systems. We build research-grade benchmarks, evaluation frameworks, and automated testing environments that push AI agents beyond demos and into practical, terminal-based workflows.

Our work includes large-scale evaluation projects such as Terminal-Bench, where we rigorously test how AI agents behave in real Linux environments. We focus on reproducibility, correctness, and robustness, combining deep systems engineering with thoughtful automation.

We’re growing quickly and are looking for engineers who enjoy hands-on problem solving, clean systems design, and real ownership.

About the Role

We are looking for a Software Engineer (Terminal Systems & Automation) to join us on a 3-month remote contract, with a strong likelihood of extension based on performance and project needs.

This role is ideal for engineers who enjoy working close to the operating system, debugging real execution environments, and building reliable automation that runs at scale.

You’ll work on designing, maintaining, and improving Terminal-Bench tasks, automation pipelines, and evaluation workflows that test AI agents in realistic Linux environments.

This is a fully remote role and starts immediately.

What You’ll Do

Design, implement, and maintain terminal-based evaluation tasks for AI benchmarking

Build and debug Bash and Python automation used in large-scale evaluation pipelines

Work extensively in Linux/Ubuntu environments, diagnosing process, filesystem, and execution issues

Create and maintain Docker-based execution environments to ensure isolation and reproducibility

Investigate flaky or failing tasks and improve their reliability, determinism, and clarity

Document task behavior, edge cases, and execution requirements

Collaborate asynchronously with a distributed engineering team to improve evaluation quality and system robustness

Who We’re Looking For

We’re looking for engineers who are curious, detail-oriented, and comfortable working independently.

Required Qualifications

Strong proficiency with Linux/Ubuntu command-line environments

Solid experience writing and debugging Bash scripts for automation

Hands-on experience with Docker (Dockerfiles, containers, volumes, execution environments)

Proficiency in Python for scripting, automation, or tooling

Good understanding of:

Processes, signals, and system behavior

Filesystems, permissions, and environment variables

Comfortable using Git in collaborative codebases

Strong debugging and problem-solving skills

Ability to work effectively in a fully remote environment

Nice to Have

Experience with Terminal-Bench, OSWorld-SFT, or similar evaluation frameworks

Background in systems programming, developer tooling, or infrastructure

Familiarity with additional languages (Go, Rust, C/C++, JavaScript)

Interest in AI systems, agent evaluation, or benchmarking

Compensation & Contract Details

$30 per completed task

Clearly scoped, engineering-focused tasks

High performers may receive:

Increased task volume

Contract extensions beyond 3 months

Long-term collaboration opportunities

Why Join Dusker AI?

Work on real systems problems, not toy projects

Flexible, remote-first work environment

Clear, measurable impact on AI reliability and evaluation

Opportunity to grow with a fast-moving, technically focused team

Merit-based opportunities for longer-term engagement