Staff/Senior Software Engineer, Agent Engineering

Helios Intelligence Platforms•New York City, NY

12d•Onsite

About The Position

Helios is building a new kind of company to solve America’s hardest problems, starting with the government interaction layer. Government shapes every consequential market, but the infrastructure connecting public institutions and private organizations remains fragmented, manual, and difficult to navigate. Helios is rebuilding that layer. Our core platform, Proxi, gives organizations the intelligence they need to understand what government is doing, why it matters, and what to do next. From that foundation, we design and deploy secure, mission-specific systems for government agencies, enterprises, and institutions operating in complex and highly regulated environments. We bring together frontier AI, deep public-sector expertise, and forward-deployed execution. Our team includes leaders and builders from the White House, U.S. Department of State, Datadog, and Microsoft. We are backed by leading institutional investors and trusted by organizations working on high-stakes problems across government and industry. MISSION Advance the agent runtime that powers Proxi’s research, analysis, and action capabilities. Our agents must decompose open-ended objectives, retrieve trustworthy evidence, use tools safely, coordinate long-running work, and produce high fidelity outputs that remain traceable to their sources. At Helios you will own the systems that make agent behavior reliable in production, including execution state, tool use, context, memory, model routing, recovery, evaluation, and security - improving Proxi’s ability to perform meaningful work over hours without losing context, exceeding their authority, silently failing, or producing unsupported conclusions. You will also advance the reasoning and memory layer built on top of the Helios Rapid Ontology System (H.R.O.S.), enabling agents to accumulate knowledge, recognize change, resolve contradictions and carry direct source context across workflows. We are looking for a senior engineer who has shipped production LLM or agent systems beyond the prototype stage. You should be comfortable diagnosing nondeterministic failures, enforcing boundaries between model judgment and deterministic software and deciding when a workflow requires autonomy, human approval, or conventional application logic. Expected areas of expertise: Production orchestration for stateful, long-running, and multi-agent workflows. Tool and action infrastructure with strong contracts, permissions, and approval boundaries. Retrieval, context engineering, evidence management, and persistent agent memory. Model-runtime engineering across providers, latency profiles, and deployment environments. Evaluation, tracing, security, and operational control of nondeterministic systems.

Requirements

Shipped production LLM or agent systems beyond the prototype stage.
Comfortable diagnosing nondeterministic failures.
Comfortable enforcing boundaries between model judgment and deterministic software.
Comfortable deciding when a workflow requires autonomy, human approval, or conventional application logic.
Production orchestration for stateful, long-running, and multi-agent workflows.
Tool and action infrastructure with strong contracts, permissions, and approval boundaries.
Retrieval, context engineering, evidence management, and persistent agent memory.
Model-runtime engineering across providers, latency profiles, and deployment environments.
Evaluation, tracing, security, and operational control of nondeterministic systems.

Nice To Haves

Experience deploying agent systems in government, defense, legal, healthcare, or other regulated environments.
Familiarity with policy and legal research where temporal accuracy, provenance, and citation precision are essential.
Experience operating self-hosted or multi-provider inference in restricted environments.
Experience building agents that create durable artifacts or perform permissioned actions on a user’s behalf.

Responsibilities

Own Proxi’s agent runtime and strengthen foundations for building new agent-driven products.
Build the systems that allow agents to own long-running work from assignment through completion. Agent activity should remain visible, steerable, and connected to the relevant project, people, evidence, deadlines, and deliverables in Action Rail.
Build the tool and action framework agents use to interact with internal systems and external services safely.
Connect agents to Proxi’s retrieval and knowledge systems so reasoning, memory, citations, and generated artifacts remain grounded in authoritative evidence.
Improve the evaluation and release discipline for agent behavior, traces and simulations to improve quality while controlling latency, cost, and risk.