Infrastructure Engineer, Agent Systems

Company Description

Hawksbill is committed to leveraging Artificial Intelligence to create a more cybersecure and ethically responsible future. Our mission is rooted in putting people first as we address global cybersecurity challenges. With a focus on innovation, we strive to develop reliable and forward-thinking solutions that prioritize ethical practices and user protection. Join us in shaping the future of technology with purpose and integrity.

Role Description

This is a full-time, on-site role based in Honolulu, HI. You'll design the platform our agents run on. That means everything between the model and the target: the sandboxed environments where agents execute code, the orchestration layer that runs thousands of rollouts in parallel, the tool surface (filesystem, network, browser, debugger, MCP integrations) that agents act through, the RL infrastructure that closes the loop.

What you'll do

• Design and operate sandboxed execution environments that host adversarial code safely and at scale
• Build the orchestration layer for parallel agent rollouts, scheduling, isolation, resource management, retries, checkpointing
• Own the tool/MCP integration surface so new capabilities ship without breaking the loop
• Build eval and replay infrastructure so every run becomes training signal and debugging artifact
• Stand up RL infrastructure (rollout collection, reward computation, async actor-learner setups) when the model work calls for it

Qualifications

• 4+ years of infrastructure or platform engineering, with experience on systems that run untrusted code, orchestrate distributed workloads, or operate at meaningful scale
• Experience at a company running serious agent or distributed-systems infrastructure (AI labs, robotics/autonomy, quantitative trading, large-scale CI or sandbox platforms)
• Deep comfort with sandboxing primitives, containers, lightweight VMs and the failure modes of each
• Experience building systems that orchestrate untrusted or semi-trusted code execution
• Strong instincts for observability and debugging in async, non-deterministic systems
• You ship. You don't over-engineer. You write the ugly version first and clean it up when the shape is clear.

Bonus

• Built RL training infrastructure (rollout/training separation, async pipelines, reward modeling)
• Security background: red team experience, CTF history, published CVEs, or prior work on offensive tooling
• Hands-on with MCP, tool-use frameworks, or custom agent harnesses
• Strong understanding of Network Security to safeguard systems and prevent potential vulnerabilities.