AI Domain Architect
Distro, Noida, Uttar Pradesh
About the job
We are hiring the architect who teaches AI what our business actually means. You will be a member of the AI Domain team, the group that owns the enterprise standards, reference architectures, approved model catalog, and governance safeguards for AI across the company. Within that team, you own the domain intelligence layer: the ontology, the agent workflow models, the fine-tuning strategy, and the intelligence pattern library that product teams build on. This is a hands-on senior individual contributor role. Patterns here are proven through working proof-of-concepts before they become standards, so you will build as much as you write.
Responsibilities Platform ontology design • Design and own the canonical ontology as a semantic layer spanning all PrismHR platforms, anchored in the central PEO and HCM platforms — entities, relationships, business rules, and terminology covering co-employment, payroll, benefits administration, workers’ compensation, onboarding, and compliance — built in partnership with the product teams who own each platform’s data model • Ground the team’s reference architectures in that ontology so retrieval, reasoning, and generated output stay consistent with how the business and its regulations actually work • Establish governance for ontology evolution as products, regulations, and customer needs change, and maintain it as a living artifact with a defined review cadence AI agent workflow modeling • Model business processes across the PrismHR platforms — PEO and HCM foremost — as agentic workflows: task decomposition, tool boundaries, decision points, escalation paths, and human-in-the-loop checkpoints mapped to the team’s risk-level requirements • Define where agents may act autonomously and where they must defer, especially in payroll, money movement, and compliance-sensitive operations • Ensure workflow models inherit the team’s embedded safeguards — validated identity claims for tenant context, audit logging, content safety — rather than reinventing them per use case