Staff AI Engineer — Agentic Systems & Evaluation

SIEL AI; United States; Remote; USD 150,000 - 210,000; Full time

Job summary

SIEL AI in the United States is seeking an experienced architect to design and own the core agent infrastructure used in client production systems. You will lead the evaluation and tracing needed to prove long-term reliability in live environments.

This hands-on role moves quickly from design to deployment, with close feedback from engineers operating the systems and a focus on production behavior as models and data evolve over time, ensuring observability, retraining triggers, and robust

Qualifications

• Strong engineering capability in agentic systems and their infrastructure.
• Experience designing orchestration, state/memory, and tool interfaces.
• Ability to build evaluations and tracing to assess reliability post-deployment.
• Focus on production behavior over time with changing models and data.
• Willingness to work closely with deploying engineers.

Responsibilities

• Build orchestration, state and memory, and tool interfaces for agentic systems.
• Design cost controls for systems that rely on models and tools.
• Develop evaluation and tracing approaches that prove system reliability.
• Detect and investigate changes in behavior after launch, including model updates.
• Support engineers deploying and debugging these systems in production.

Skills

Agentic systems engineering

Orchestration design

State/memory design

Tool interfaces

Evaluations and tracing

Production behavior monitoring

Team collaboration

Job description

SIEL AI in the United States is seeking an experienced architect to design and own the core agent infrastructure used in client production systems. You will lead the evaluation and tracing needed to prove long-term reliability in live environments.

This hands-on role moves quickly from design to deployment, with close feedback from engineers operating the systems and a focus on production behavior as models and data evolve over time, ensuring observability, retraining triggers, and robust