Senior AI Engineer

Narvar is Growing! We’re building Navi — Narvar’s agentic AI that automates post-purchase resolution for the world’s leading retailers. Hundreds of millions of consumers interact with Narvar every year. Navi is our agentic AI that resolves delivery issues, returns, and refunds through natural conversation — powered by IRIS and 74 billion consumer touchpoints. 

We're looking for senior AI engineers to own this system end-to-end: architecture, model selection, production operations. You'll help decide what gets built and how.

Day-to-day

• Design and build conversational AI agents for returns, claims, and customer service experiences
• Own agent systems from architecture → implementation → evaluation → production operations
• Build RAG / context graph retrieval pipelines that ground agent responses in real company and customer data
• Design agent orchestration for multi-step workflows that interact with identity, risk, order, and loyalty systems
• Create evaluation frameworks to measure task completion, accuracy, safety, and user satisfaction
• Implement guardrails and safety mechanisms — content moderation, hallucination detection, graceful fallbacks
• Integrate conversational experiences across web, mobile, SMS, and email channels
• Make real decisions around prompt design, model selection, latency/cost/quality tradeoffs, and failure modes
• Collaborate with product, design, and ML teams to build systems that are technically sound and product-aware

What We’re Looking For

We care more about judgment and ownership than credentials.

You’re likely a strong fit if you:

• Have shipped conversational AI or agent-based systems used by real users in production
• Have built production systems on top of LLM APIs and agent frameworks — not just prompt playgrounds, but real integrations involving tool orchestration, context management, and reliability at scale
• Have a point of view on model selection tradeoffs — when to use frontier APIs vs. open-weight models (Qwen, Llama, Mistral), and understand the cost, latency, privacy, and capability tradeoffs of each
• Understand prompt engineering beyond basics: structured outputs, few-shot learning, chain-of-thought, tool calling
• Have built context graph pipelines that go beyond naive retrieval — entity resolution, relationship modeling, and dynamic context assembly from structured and unstructured data
• Have designed agent architectures that use function calling, tool execution, or multi-step reasoning