Multi-variate time-series system. Predicted quarterly revenue within 3.2% accuracy. Used by finance for resource allocation.

Problem Space

Built a multi-variate time-series forecasting system that predicts quarterly revenue with 3.2% accuracy. The system ingests data from multiple sources including CRM, billing, and market indicators to produce reliable forward-looking estimates.

Solution Approach

The pipeline runs on Airflow with automated retraining cycles. Feature engineering includes lag variables, rolling statistics, and external economic indicators. Model selection is automated through a tournament approach comparing Prophet, LightGBM, and ensemble methods.

Homepage hero section

Impact

The finance team uses this daily for resource allocation and headcount planning. Before this system, forecasts were spreadsheet-based and routinely off by 15-20%.