Senior Engineer, Production LLM/Agent Runtime

Helios; New York, Northern (NY, KY); Hybrid; USD 170,000 - 210,000; Full time

Job summary

Helios, based in New York City, is seeking a Staff/Senior Software Engineer for Agent Engineering to advance the Proxi agent runtime and related systems on-site in SoHo. You will own production-grade agent execution, memory, and tool-use frameworks in a fast-paced startup environment.

You will diagnose nondeterministic failures, reason about autonomy versus human approval, and collaborate with the founding team to ship reliable agent capabilities across regulated environments.

Qualifications

• Senior engineer who has shipped production LLM or agent systems beyond prototype stage.
• Comfortable diagnosing nondeterministic failures and enforcing boundaries between model judgment and deterministic software.
• Experience with stateful, long‑running, multi‑agent workflows and secure, auditable tool usage.

Responsibilities

• Own Proxi’s agent runtime and foundations for new agent‑driven products.
• Build systems to manage long‑running work from assignment to completion with visible, steerable activity connected to projects, people, evidence, deadlines, and deliverables.
• Develop the tool and action framework used by agents to safely interact with internal and external services.
• Integrate agents with Proxi’s retrieval and knowledge systems to ground reasoning, memory, citations, and artifacts in authoritative evidence.
• Improve evaluation and release discipline for agent behavior, traces, simulations, latency, cost, and risk.

Skills

Agent runtime

Tool framework

Evidence management

Model-runtime engineering

Evaluation & tracing

Job description

Helios, based in New York City, is seeking a Staff/Senior Software Engineer for Agent Engineering to advance the Proxi agent runtime and related systems on-site in SoHo. You will own production-grade agent execution, memory, and tool-use frameworks in a fast-paced startup environment.

You will diagnose nondeterministic failures, reason about autonomy versus human approval, and collaborate with the founding team to ship reliable agent capabilities across regulated environments.