Machine learning–powered demand forecasting and inventory optimization for retail and fashion. This project implements a full pipeline on Databricks: multi-modal data (time series, images, tabular), uncertainty-aware forecasting with conformal prediction, and MLOps with drift monitoring and inventory dashboards.
- Overview
- Author & Contact
- Features
- Technology Stack
- Project Structure
- Data Pipeline
- Dashboards
- Usage & License
- Disclaimer
This repository demonstrates end-to-end demand forecasting and inventory optimization using Databricks. It uses real-world fashion industry data from Visuelle 2.0.
More on the dataset: Visuelle – Humatics Lab.
What this project covers:
- Structured streaming and processing for multi-modal data (images, time series, and tabular)
- Bronze → Silver → Gold data pipeline and MLOps for training and evaluating ML models
- Multi-modal forecasting (LightGBM + Nixtla MLForecast) with image embeddings
- Uncertainty quantification via conformal prediction (MAPIE)
- MLOps dashboards for model monitoring, error analysis, and inventory optimization
KuchikiRenji
| KuchikiRenji@outlook.com | |
| GitHub | github.com/KuchikiRenji |
| Discord | kuchiki_renji |
| Feature | Description |
|---|---|
| Multi-modal data | Sales, price, weather, Google Trends, customer stats, and image embeddings (imgbeddings) |
| Medallion architecture | Bronze (raw) → Silver (cleaned) → Gold (forecast-ready and inventory) |
| Demand forecasting | 2-weeks-in → 1-week-ahead forecasts with LightGBM (Tweedie) and Nixtla MLForecast |
| Uncertainty | Prediction intervals via MAPIE (conformal prediction) |
| MLOps | Evidently AI for data/target drift and regression quality; MLflow for model registry and serving |
| Inventory | Gold tables and dashboards for restock and inventory optimization |
- Platform: Databricks (Spark, Delta, Unity Catalog)
- Forecasting: Nixtla (MLForecast), LightGBM
- Uncertainty: MAPIE (conformal prediction)
- Image embeddings: imgbeddings
- Monitoring: Evidently AI
- Model registry & serving: MLflow, Databricks Unity Catalog
├── src/
│ ├── data_preprocessing/
│ │ ├── bronze/ # Raw ingestion (sales, price, weather, gtrends, customer, clothes, restock)
│ │ ├── silver/ # Cleaned and joined tables
│ │ └── gold/ # Forecast inputs, forecasted outputs, inventory, restock, image embeddings
│ ├── model_training/ # Nixtla + LightGBM + MAPIE training notebook
│ ├── model_evaluating/ # Evidently AI drift and regression monitoring
│ └── model_serving/ # MLflow model registration and serving API
├── LICENSE
└── README.md
End-to-end flow from raw data to forecasts and inventory:
This project is for educational and experimental use and is non-commercial. Usage must comply with the license of the data source (Visuelle 2.0).
This project is licensed under the MIT License. See the LICENSE file for details.
This work is not affiliated with or endorsed by any third parties. The author assumes no responsibility for the accuracy, completeness, or consequences of using the content provided. Use of the materials in this repository is at your own risk.
Demand forecasting · Inventory optimization · Time series forecasting · Retail analytics · Fashion demand · Machine learning · MLOps · Databricks · LightGBM · Conformal prediction · MAPIE · Nixtla · Evidently AI · Multi-modal ML · PySpark · Delta Lake