Systems & Research Engineer

Location: Remote (global)  |  Comp: $220K to $300K TC  |  Type: Full-time  |  Openings: 4

About the Company

A well-funded applied AI lab building fraud detection, voice-agent, and document-processing systems for a large, traditionally underserved industry. Remote-first team with a presence in San Francisco. Small, high-caliber engineering group with a strong bar for independent thinking and daily use of coding agents.

About the Role

Own the infrastructure, performance, and AI-systems research behind the company's core products, including document ingestion and inference serving. This role blends performance engineering, benchmark-driven research, and distributed-systems tradeoff analysis, with close collaboration alongside applied engineers.

What You'll Own

• Core-system performance across inference, serving, and retrieval
• Structured AI research: hypothesis, benchmark design, data-driven recommendations
• Serving tradeoffs around concurrency, latency, and cost
• Clear communication of system constraints to the broader team
• Heavy daily use of coding agents as part of the workflow

What We're Looking For

• Demonstrated performance engineering on an AI model-system mechanism (inference, serving, evaluation, retrieval, speech, or agent infrastructure). Distributed systems work with no model in the loop does not qualify.
• Comfort explaining a hypothesis, the alternatives considered, and how bias in data or benchmarks was addressed
• Fluent, daily use of coding agents, assessed directly in the interview process
• Visible side projects, even small or incomplete ones, showing independent curiosity
• Bonus: AI/ML research background or a physics/quantitative-science background moving into software

Compensation Note

Senior-level guidance is roughly $240K to $300K TC. Requests meaningfully above that need an exceptional, clearly communicated case.

To Apply: Submit your resume along with a short note on the most technically interesting system you've built or improved and what the actual bottleneck was.