Challenge
Retail clients over-ordered slow movers and ran out of bestsellers. Forecasts lived in spreadsheets that took a week to update.
Solution
- Built forecasting pipelines combining gradient-boosted models with seasonality and promotion signals.
- Created a natural-language query layer so planners can ask questions in plain English.
- Deployed on Kubernetes with nightly retraining and drift monitoring.
- Designed a PostgreSQL warehouse with partitioned fact tables for fast analytics.
Result
Stockouts fell 31%, excess inventory dropped 22%, and forecasts now refresh daily instead of weekly.
Tech stack
- Python
- Claude
- PostgreSQL
- Kubernetes
- AWS
- Next.js
Services: AI & Automation · DevOps & Cloud Infrastructure · Database Architecture
