DemandIQ
Retail Demand Forecasting & Replenishment Engine
Forecasts next-week sales for every SKU in every store — then turns the forecast into a reorder quantity and a risk alert.
- Context
- Capstone-I · IIT Patna
- Role
- Led model training
- Dataset
- Walmart M5
- Year
- Mar 2026
01 Problem
A forecast is only useful if it changes what’s on the shelf. DemandIQ goes from raw sales history to a reorder quantity and a risk flag a store manager can act on.
02 Approach
Ingest and clean
Raw sales, calendar and price data from the Walmart M5 dataset flow through ingestion and cleaning stages into a feature store.
Engineer features
Lag features at 7, 14 and 28 days, rolling statistics, price deltas and holiday flags.
Forecast with two models
Prophet captures seasonality; XGBoost captures the more complex patterns. Models are compared per SKU on MAE, with experiments tracked in MLflow.
Decide the reorder
Safety stock = Z × σ × √lead time at a 95% service level. Inventory is classified into LOW / MED / HIGH risk.
Serve and alert
FastAPI REST endpoints, a Streamlit dashboard, and real-time Telegram alerts for HIGH-risk inventory.
03 Results
- Pipeline layers
- 6
- Day lag features
- 7·14·28
- Service level target
- 95%
- Forecast models
- 2
- XGBoost outperformed Prophet on most SKUs by MAE.
- A 6-layer modular pipeline, from ingestion to alerting.
- Led model training for the team.
04 Stack
- Prophet
- XGBoost
- FastAPI
- Streamlit
- MLflow