Varun

Business Intelligence Engineer

Amazon • Fulltime

Sep 2024 - Jan 2026 • 1 yr 4 mos

1. RDR Analytics, Pipeline Optimization & AI Agent Development • Reduced RDR (Repeat Defects Reduction) metric reporting lag by 79% (14 → 3 days) and query execution time by 64% (7 → 2.5 hrs) through defect pattern analysis and Spark SQL pipeline redesign. • Developed RDR dashboards with 20+ dimensions; uncovered that pet-presence addresses contribute to 23% of defects, leading to pet-aware delivery instructions and targeted messaging. • Identified SDPP as a key false positive in the RDR metric during phased U.S. rollout; removed SDPP-related defects from reporting, eliminating a∼20 DPMO distortion and enhancing reliability. • Partnered with Data Engineering to build Amazon’s internal Insight Generator agent (Apostrophe) leveraging Claude Sonnet 4.5; owned end-to-end BI ETL pipeline and system prompts, cutting manual effort by∼50%. 2. Delivery Hours Model Performance & A/B Experimentation • Analyzed v1 and v1.5 delivery hours science models that expanded customer business hours using heuristics and delivery history data. • Conducted an A/B test evaluating first-attempt delivery failure and time window compliance; performed statistical sig- nificance testing using Permutation test to rule out randomness. • Engineered targeted deep dives to quantify impact across aggressive and conservative variants; the conservative variant improved closed accuracy by 77 bps (p-value = 0.0063) and delivery failure by 200 DPMO (p-value=0.07, near significance), enabling 100% U.S. rollout.