Software Engineer – AI Research & Evaluation (US-based)

What You'll Do7

• Curate code samples and craft solutions in Python, C/C++, Rust, Go, Java, and JavaScript (including ReactJS) for model training.
• Review and improve AI-generated code, focusing on systems-level correctness, speed, and dependability.
• Build agents that automatically check the quality of infrastructure and systems code and flag recurring error patterns.
• Form hypotheses about stages of the software engineering cycle (prototyping, architecture, API design, production, launch, experiments, monitoring, maintenance) and test model capabilities on each.
• Design verification mechanisms that can automatically confirm whether a solution to an engineering task is correct.
• Work with cross-functional teams to refine AI-driven coding tools against industry performance benchmarks.
• Write clear, structured rationales for evaluation decisions and share findings with researchers.

Requirements6

• At least 3 years of software engineering experience.
• Strong background in systems programming, infrastructure, or backend development using languages such as Python, C/C++, Rust, and Go.
• Proven ability to build and deploy scalable, production-grade software with modern languages and tools.
• Deep knowledge of software architecture, design, debugging, and code quality assessment.
• Excellent written and verbal communication skills, especially for writing clear evaluation rationales.
• Must be based in the United States.

Preferred Qualifications2

• Experience at frontier AI companies such as OpenAI, NVIDIA, Databricks, Palantir, or Snowflake.
• Graduation from leading programs like Harvard, Columbia, Princeton, Yale, or University of Pennsylvania.

Who Should Apply

Engineers with 3+ years of systems, infrastructure, or backend experience who have worked with Python, C/C++, Rust, or Go and want to shape how AI models handle real software engineering tasks. The role fits someone who can write production-grade code, critique it, and explain their reasoning in clear written form.