AI/MLForecasting/Time-SeriesMLOps/Deployment
End-to-End Demand Forecasting and Inventory Optimization
Multi-modal ML pipeline with uncertainty-aware forecasting
End-to-end demand forecasting and inventory optimization with Databricks (Nov 2024 – Jan 2025) - multi-modal ML (time series, images, tabular), LightGBM, MAPIE conformal prediction and MLOps for retail and fashion. Implements full pipeline on Databricks with uncertainty-aware forecasting, drift monitoring, and inventory dashboards.
Tech Stack
PythonDatabricksPySparkLightGBMMLflowDelta LakeEvidently AIGit/GitHub
Technical Highlights
- Multi-modal ML pipeline: time series, images (imgbeddings), and tabular data on Databricks
- Forecasting with Nixtla MLForecast and LightGBM; uncertainty quantification via MAPIE conformal prediction
- MLOps: Evidently AI for drift monitoring, MLflow for model registry and serving, Unity Catalog
- Bronze-Silver-Gold data architecture with Delta Lake for sales, price, weather, trends, and inventory
Challenges & Solutions
- Uncertainty in forecasts → Implemented MAPIE conformal prediction for prediction intervals
- Multi-modal data integration → Built unified pipeline combining time series, images, and tabular features
Gallery