Description

Aza is Amazon's primary internal AI assistant, serving 1.5M+ employees globally across HR, IT, pay, benefits, and workplace domains. We're not wrapping an LLM with a chat UI — we're building a full-stack agentic platform: 60+ production sub-agents, multi-turn orchestration, real-time skill routing, MCP-based tool integrations, and enterprise-grade evaluation systems.
We're looking for an SDE-II who wants to build AI-native systems from the ground up.

What makes this role unique:
* You'll build the agent, not just use one. Aza 2.0 is built on Strands SDK with dedicated ECS Fargate compute, custom orchestration, and a plugin architecture spanning 60+ sub-agents. You'll own real components of this system.
* Greenfield Gen AI engineering. We're actively shipping: agentic inline workflows, voice-enabled agents, RAG pipelines, progressive skill loading, and a self-service developer platform for 50+ away teams building on Aza.
* Complex, high-stakes domain. Aza handles questions about pay, benefits, leave, HR policies — across every jurisdiction Amazon operates in. Correctness matters. This is where AI reasoning meets real-world constraints.
* Scale that doesn't exist elsewhere. Millions of conversations monthly. Sub-3-second response latency. Production reliability with Sr. PE-level architectural standards.
* Visibility and mentorship. SVP-level visibility, Sr. PE driving architectural vision, and a fast-shipping team where engineers own end-to-end features in production.

Key job responsibilities
1. Design and build agentic systems: orchestration, skills, sub-agents, MCP tools, evaluation frameworks
2. Ship features end-to-end from design through production deployment
3. Work across the stack: LLM integration, distributed systems, real-time APIs, infrastructure-as-code
4. Collaborate with Sr. PE, PEs and Sr. engineers on architecture decisions that shape the platform's future