Senior Applied Scientist, Amazon Global Data Center Ops Central Insight and Analytics Team

We are looking for an seasoned Applied Scientist to design, build, and deploy the ML/AI models that power our decision intelligence platform. You will work at the intersection of causal inference, time-series forecasting, anomaly detection, and LLM-based reasoning - all applied to real operational problems with measurable business impact.

Key Job Responsibilities

Decision Intelligence Models

• Causal inference & root cause analysis: Build models that decompose fleet-wide metric movements into root causes, distinguishing correlation from causation across operational dimensions (site, service, failure mode, time)
• Dose-response modeling: Develop models that learn the quantitative relationship between intervention intensity and outcome magnitude
• Forecasting & projection: Build time-series models that project metric trajectories under different intervention scenarios, enabling "if we do X, expect Y by date Z" recommendations
• Anomaly detection & trend identification: Develop multi-variate anomaly detection that distinguishes signal from noise in noisy operational data, and identifies emerging patterns before they become crises
• Confidence calibration: Build and maintain calibrated confidence scores for recommendations, ensuring the system knows what it knows and what it doesn"t
• Outcome attribution: Design experiments and causal methods to measure the true impact of interventions

LLM Integration & Reasoning

• Structured reasoning: Design LLM prompting architectures that reliably transform operational data into executive-quality narrative summaries, decision framings, and recommendation rationales
• LLM evaluation: Build evaluation frameworks that measure LLM output quality (accuracy, actionability, calibration) and detect degradation over time
• RAG systems: Design retrieval-augmented generation systems that ground LLM outputs in operational data, historical playbooks, and institutional knowledge
• Progressive autonomy: Design the trust-calibration system where AI gradually earns expanded authority based on demonstrated accuracy over time

Research & Production

• End-to-end ownership: Take models from research through production deployment - you ship, you monitor, you iterate
• Experimentation: Design A/B tests and quasi-experiments to validate model improvements and measure business impact
• Stakeholder communication: Translate complex scientific results into actionable insights for non-technical senior leaders