Head of Applied AI

NIURAL Head of Applied AI

New York City
· Onsite
· Full-Time
· Reports to CEO & CTO

Niural is the AI-native platform that unifies payroll, compliance, HR, and financial operations across 150+ countries. Our AI agent layer, EMMA, doesn't sit on top of the product, it is the product. We process high‑stakes, multi‑jurisdictional payroll and compliance data where a single error can mean regulatory penalties, missed paychecks, or broken trust.

We're hiring a Head of Applied AI to own the entire AI engineering function. This is a hands‑on technical leadership role for someone who ships production systems, not someone who manages from a distance. You will design, build, and iterate on the AI systems that make Niural's platform intelligent, and you will set the engineering standard for the team that grows around you.

Backed by Marathon, M13, and Inspired Capital.

• Multi-jurisdictional document understanding across 150+ countries, each with distinct payroll rules, tax codes, and employment regulations, in dozens of languages.

• Structured data extraction from payroll and compliance inputs where accuracy isn't aspirational, it's a regulatory requirement. You will build systems where the cost of a hallucination is a compliance violation, not a bad search result.

• Agentic workflows (EMMA) that need to execute autonomously on financial transactions while remaining explainable and auditable. This isn't a chatbot that just answers questions, it's autonomous payroll execution.

• Real‑time anomaly detection on financial data across currencies, jurisdictions, and tax regimes simultaneously.

If you want to do serious applied AI work in a domain where the output actually matters, where model performance translates directly to whether people get paid correctly, this is the role.

You report to the CEO and CTO. Your focus is on delivering, getting AI systems into production, iterating on model performance, and building engineering practices that scale.

• Production AI Systems. Design, build, and ship LLM‑powered features end‑to‑end: document understanding, structured data extraction from payroll and compliance inputs, multi‑step reasoning workflows, and intelligent automation across the Niural platform. You write and review code. You deploy to production. You own the outcome.

• Evaluation & Reliability. Construct rigorous evaluation frameworks with ground‑truth datasets, regression tracking, and clear production‑readiness criteria. In regulated financial and employment contexts, works most of the time is not good enough. Define what good means, measure it, and hold the bar.

• AI Infrastructure. Build and own the core stack: RAG pipelines, vector stores, embedding systems, fine‑tuning workflows, model serving, and observability tooling. Make principled build‑vs‑integrate decisions based on accuracy, cost, latency, and data privacy.

• Responsible AI in Practice. Implement hallucination controls, confidence scoring, human‑in‑the‑loop review flows, and audit trails for model‑driven decisions. This is a domain where responsible AI isn't a slide deck, it's an engineering requirement with regulatory consequences.

• Cross‑Functional Translation. Work with product and engineering to translate business requirements into concrete model specifications, data requirements, and acceptance criteria. Bridge the gap between we need AI to do X and a shipped, measurable feature.

• Team & Engineering Culture. Set the engineering foundations, code standards, review processes, documentation practices, that will support a growing AI team. You'll be the first dedicated AI hire; the team grows around the standard you set.

We care about what you've built and shipped, not where you trained. Show us the production systems, the eval frameworks, the hard tradeoffs you've made.

• 6+ years hands‑on in ML or AI engineering, with at least 2 years building and shipping production LLM systems, not notebooks, not prototypes, production.

• Strong Python engineering skills. You write production‑quality code that other engineers can review, test, and extend.

• Direct experience with RAG architectures, prompt…