An end-to-end operations analytics pipeline examining delivery delays, refund patterns, category performance, and order-level refund risk using a reproducible synthetic dataset.
E-commerce platforms frequently encounter delivery delays, perishable fulfillment bottlenecks, and escalating refund requests. Without order-level analysis, operations teams cannot isolate whether customer complaints stem from courier SLA breaches, category handling issues, or service ratings.
Engineered an auditable, reproducible analytics pipeline featuring a seeded synthetic dataset generator with 21 schema validation checks, an ETL feature-engineering layer in Python/pandas, a PostgreSQL SQL analytical layer with window functions, dual refund-risk scoring (explainable SQL rules vs. trained machine learning), and an interactive multi-tab Streamlit dashboard.
| Component Layer | Role & Implementation |
|---|---|
| Seeded Data Generator | Generates 100k reproducible orders with 21 automated validation checks (validate_data.py) |
| ETL & Feature Pipeline | Cleans timestamps, derives promised-vs-actual delays, and extracts ML features via pandas |
| PostgreSQL SQL Layer | Analytical queries using CTEs, window functions, and FILTER clauses reproducing notebook metrics |
| Predictive Modeling | Stratified 80/20 train/test split with 5-fold cross-validation (Logistic Regression & Random Forest) |
| Streamlit Dashboard | Interactive 4-tab operational UI with platform, category, and date range filters |
Get in Touch
Tell me about your agency challenge, spreadsheet bottlenecks, or workflow needs. I will review the details and suggest a practical approach.
Prefer direct communication over the form? Send an email or connect via WhatsApp: