Staff Software Engineer

Role summary

The Staff Software Engineer (IC4) on HR Service Delivery designs, builds, ships, and operates capabilities whose core behavior is model-driven rather than explicitly authored: agentic and conversational experiences that interpret an employee's, manager's, or HR agent's intent, reason over employee profile, case, catalog, policy, and knowledge context, invoke tools, and act on the user's behalf across the employee lifecycle.

This is not a machine learning or AI research role; the engineer does not train foundation models. It is also distinct from traditional full-stack engineering, where systems follow deterministic logic rather than selecting execution paths at runtime. Two consequences shape the work. First, the most important logic often lives in natural language: instructions, prompts, tool descriptions, guardrails, escalation rules, all of which must be engineered, versioned, and reviewed with the same discipline as code. Second, because behavior is probabilistic, correctness is established by measuring behavior at scale rather than by asserting fixed outputs, making automated evaluation a first-class engineering activity rather than a quality-assurance afterthought.

HR sharpens both points along two independent axes.

Accuracy. An agent answering on payroll, benefits, leave, or a lifecycle event touches statutory entitlement, jurisdiction-specific policy, and an employee's pay. A confidently wrong output is not a bad answer; it is a missed enrollment window or an incorrect leave balance acted on in good faith.

Audience. HR data is among the most sensitive on the platform, and correctness of content is not sufficient. An answer grounded in a record or knowledge article the requestor is not entitled to read is a data exposure even when every fact in it is true, and manager-scope and employee-scope views of the same question have different correct answers. Access boundaries have to be enforced in the retrieval and tool layer rather than requested of the model.

At IC4 the engineer owns AI design decisions across the domain, not within a single feature, and owns the correctness of what ships whether a person or an agent produced it.

What you do Build AI-native capability across the employee lifecycle

Design and ship features built around agentic behavior — intent interpretation, multi-step reasoning, tool invocation, and action on the user's behalf — together with the data models, integrations, and channels that make them usable in production. In HRSD this spans guided service selection and intake, natural-language case creation and enrichment, conversational case status and in-flight change, tiered resolution of payroll and benefits inquiries, case triage, routing, and deflection, eligibility and entitlement inquiry, and agent-driven execution of onboarding, transfer, and offboarding lifecycle events.

Design AI-driven autonomous workflows

Decompose HR processes into the steps and decision points an agent can execute: determining where autonomy is appropriate, where a checkpoint with a person is required, and how exceptions, retries, and hand-back are handled. Anything that changes pay, employment status, or a restricted record, and anything touching employee relations or investigation, needs a human decision point by construction. The design must make that distinction structural rather than advisory.

Author and maintain agentic instructions as engineering artifacts

Write, structure, and version the system instructions, role definitions, tool descriptions, guardrails, and escalation paths that govern agent behavior in the domain, under code review, source control, and regression coverage. Own the shared instruction and tool-description surface that adjacent teams build against. Because HRSD ships as product, customers configure, extend, and override that surface on their own instances: treat it as public API, with upgrade-safe extension points and versioning discipline to match.

Build automated evaluation and test non-deterministic behavior

Design and operate the evaluation that makes change safe: golden datasets, multi-turn conversation suites, model-as-judge scoring calibrated to human review, CI gates, and drift detection, plus adversarial, jailbreak, grounding, and tool-selection testing. Extend that coverage to the HR-specific failure classes — access-boundary violations in retrieval and citation, PII leakage across scopes, and jurisdictional and policy-variant correctness — and keep evaluation meaningful across customer configurations rather than against a single reference dataset. Own the resolution, containment, and quality metrics the domain is measured on, including whether they are instrumented correctly in the first place.

Design conversational experiences across channels

Build experiences that hold context across turns, hand off cleanly between automated and live HR agents, and behave consistently across employee-facing portals, chat shells, workplace messaging clients, agent workspace, and voice — accounting for what voice imposes: latency budgets, barge-in, speech recognition error on names and plan terminology, disambiguation, and explicit confirmation before consequential actions.

Specify precisely and direct AI coding agents

Convert requirements into testable specifications with explicit scope, constraints, non-goals, and acceptance criteria; decompose work into agent-sized tasks; supervise several workstreams in parallel; and review agent output for correctness, spec adherence, security, and maintainability. You own the result regardless of what produced it.

Own quality, safety, and reliability in production

Monitor conversation quality, containment, hallucination rate, tool-selection error, and unsafe or unauthorized action. Defend against prompt injection and data leakage across integration surfaces, including the paths where user-supplied content — case notes, inbound email, attachments, authored knowledge — enters agent context. Maintain reasoning-trace observability and model rollback mechanisms, and feed production failures back into specifications and evaluation sets. Because HR conversation content is itself restricted, design that observability to be diagnosable without exposing what was said. Lead root-cause analysis when agentic behavior deviates from intent, and hold the line between a genuine model failure and a platform or configuration failure presenting as one.

Ground it in solid full-stack delivery

Build the application, APIs, data models, and integrations around these capabilities: front-end experiences for employees, managers, and HR agents, server-side logic, and the connections to HCM, payroll, benefits, identity, knowledge, and the adjacent service domains HR cases cross into — with the CI/CD, observability, and upgrade-safe extensibility expected of production software.

Collaborate across product, design, and engineering

Partner with product managers, designers, conversation designers, HR domain and compliance partners, and engineers to define success criteria and communicate capability and risk clearly. Mentor IC1 to IC3 engineers, and raise the team's practices around instruction authoring, evaluation, and accountable agent use.

Required experience and skills

• Experience in leveraging or critically thinking about how to integrate AI into work processes, decision-making, or problem-solving. This may include using AI-powered tools, automating workflows, analyzing AI-driven insights, or exploring AI's potential impact on the function or industry.

• 9+ years software engineering experience (backend, frontend, or full stack) 

          Strong JavaScript/Node.js, React and API integration skills  

          Automated testing (unit + integration) and CI/CD pipeline experience 

          Solid understanding of data modelling and query optimization 

• Agentic delivery experience: 

           Hands-on experience with AI/GenAI or ML-driven features (LLM prompting, NLU,  
           classification models, or similar) 

          Hands-on experience with LLM-integrated features: prompt design, context injection, output tuning,                    guardrails.  

          Experience with NLU/intent classification or conversational AI systems 

          Familiarity with ML-driven automation (classification, clustering, recommendation  
          systems) 

• Production AI integration. Experience integrating large language model APIs and retrieval-grounded features, including agent orchestration, tool and function calling, and structured output enforcement. 

• Applied machine learning literacy. A working command of the concepts that govern how these systems behave — evaluation, embeddings, and the probabilistic output and failure modes of modern models — sufficient to reason about, debug, and verify model-driven behaviour in production. 

• Accountable use of AI coding agents. Current, effective use of AI coding assistants and agents with evidence of accountable delivery: precise specification, critical review of generated output, and verification harnesses. 

• Operational experience. Hands-on CI/CD, containerized workloads, and observability experience, plus direct on-call and incident-command experience with customer-facing systems. 

• Mentorship. Demonstrated mentorship of less-experienced engineers and a record of raising quality through code review. 

• Education. Bachelor's degree in computer science, software engineering, or a related technical field, or equivalent practical experience. Advanced degrees are a plus but not a substitute for a record of shipping reliable AI-native applications.

Preferred experience 

• HR domain depth. HR case management, the employee lifecycle, or payroll, benefits, leave, and absence, at a depth sufficient to challenge a requirement rather than only implement it — including why the same policy question resolves differently by jurisdiction, employment type, or plan year. 

• HCM and downstream integration. Integration with core HR, payroll, and benefits systems of record, including the reconciliation and eventual-consistency problems that come with treating an external system as the authority on employment data. 

• Extensible product engineering. Building capability that customers configure, extend, and override on their own instances, where instruction and tool surfaces are versioned contracts rather than internal implementation. 

• Evaluation and observability tooling. Evaluation frameworks, prompt and instruction management tooling, tracing for model-driven applications, and analysis of production transcripts at scale, particularly where transcript content is itself access-restricted. 

• Conversational channel breadth. Employee self-service portals, virtual agent or chat shells, workplace messaging clients, and voice, including handoff between automated and live agents. 

• Conversation design partnership. Working alongside conversation or content designers on dialogue flow, tone, and error-recovery design — with attention to how an HR answer is worded when the subject is the employee's own pay, health coverage, or employment status. 

• Forward deployed delivery. Building against a customer's data, integrations, and channels, and tuning instructions and evaluation sets in their environment. 

• Now Platform depth. Scoped applications, ACLs and platform security rules, Flow Designer, UI Builder, Automated Test Framework, and upgrade-safe extension patterns.