Software Dev Engineer II, Long Term Planning and Forecasting
Amazon.com
Build the AI system that decides how Amazon plans. PIXIE is a new agentic AI platform that routes natural-language planning questions to specialist agents, generates trustworthy data-grounded answers, and automates business narratives — all serving analysts across Amazon's supply chain and finance organizations. We're hiring an SDE-II to own core platform subsystems in a greenfield, high-ambiguity environment where your architectural decisions ship to production within weeks.
- You'll design and build production systems at the intersection of LLMs, multi-agent orchestration, and large-scale data retrieval
- You'll work on novel problems: agentic routing with confidence calibration, stochastic evaluation, and authorization-aware knowledge retrieval
- You'll ship visible impact — PIXIE is LTPF's flagship AI investment with VP-level visibility and direct analyst adoption metrics
- Tight engineering team (8 SDEs), high autonomy, rapid iteration cycles
Key job responsibilities
- Design, build, and operate core platform services for PIXIE's agentic AI system — including agent runtime, caching/cost-latency optimization, observability, and evaluation infrastructure
- Build and maintain the tool/agent registry that enables new capabilities to be discovered and invoked without redeployment
- Develop automated evaluation and regression harnesses for non-deterministic LLM outputs — hallucination detection, drift monitoring, and faithfulness scoring
- Implement authorization-aware retrieval ensuring users never see content they aren't entitled to
- Collaborate with Scientists on grounding methodology, causal analytics productionization, and retrieval-quality optimization
- Operate what you build — on-call rotation, runbooks, operational dashboards, and continuous improvement of system reliability
- Mentor teammates on system design, code quality, and operational best practices
About the team
Long Term Planning and Forecasting (LTPF) owns the models, systems, and tools that power Amazon's long-range demand planning — the forecasts that drive billions in inventory, capacity, and capital decisions. PIXIE is LTPF's AI-native platform: a conversational intelligence layer that lets analysts ask questions in natural language and get trustworthy, data-grounded answers backed by Amazon's planning data.
We're a small engineering team building from zero to production. Our stack spans multi-agent orchestration, retrieval-augmented generation, causal inference, and automated business narrative generation. We operate like a startup inside Amazon — fast iteration, high ownership, direct customer feedback loops — with the backing of Amazon-scale data and infrastructure.
We value builders who thrive in ambiguity, care deeply about correctness (especially when LLMs are involved), and want to shape the architecture of a system from its earliest days. If you've been looking for a role where you can combine distributed systems engineering with the frontier of applied AI — and see your work used by thousands of planners across Amazon — this is it.
- You'll design and build production systems at the intersection of LLMs, multi-agent orchestration, and large-scale data retrieval
- You'll work on novel problems: agentic routing with confidence calibration, stochastic evaluation, and authorization-aware knowledge retrieval
- You'll ship visible impact — PIXIE is LTPF's flagship AI investment with VP-level visibility and direct analyst adoption metrics
- Tight engineering team (8 SDEs), high autonomy, rapid iteration cycles
Key job responsibilities
- Design, build, and operate core platform services for PIXIE's agentic AI system — including agent runtime, caching/cost-latency optimization, observability, and evaluation infrastructure
- Build and maintain the tool/agent registry that enables new capabilities to be discovered and invoked without redeployment
- Develop automated evaluation and regression harnesses for non-deterministic LLM outputs — hallucination detection, drift monitoring, and faithfulness scoring
- Implement authorization-aware retrieval ensuring users never see content they aren't entitled to
- Collaborate with Scientists on grounding methodology, causal analytics productionization, and retrieval-quality optimization
- Operate what you build — on-call rotation, runbooks, operational dashboards, and continuous improvement of system reliability
- Mentor teammates on system design, code quality, and operational best practices
About the team
Long Term Planning and Forecasting (LTPF) owns the models, systems, and tools that power Amazon's long-range demand planning — the forecasts that drive billions in inventory, capacity, and capital decisions. PIXIE is LTPF's AI-native platform: a conversational intelligence layer that lets analysts ask questions in natural language and get trustworthy, data-grounded answers backed by Amazon's planning data.
We're a small engineering team building from zero to production. Our stack spans multi-agent orchestration, retrieval-augmented generation, causal inference, and automated business narrative generation. We operate like a startup inside Amazon — fast iteration, high ownership, direct customer feedback loops — with the backing of Amazon-scale data and infrastructure.
We value builders who thrive in ambiguity, care deeply about correctness (especially when LLMs are involved), and want to shape the architecture of a system from its earliest days. If you've been looking for a role where you can combine distributed systems engineering with the frontier of applied AI — and see your work used by thousands of planners across Amazon — this is it.
Confirmar seu email: Enviar Email