The Problem
Retail demand is volatile and hard to predict. Inaccurate forecasts lead to stockouts, overstocking, and lost sales across 1000+ stores.
RETAIL · GENAI · DEMAND FORECASTING
GenAI Retail Forecast
Demand forecasting and retail intelligence platform for Rossmann with real-time dashboards, GenAI narratives, and automated alerts.
GitHub ↗Live Demo ↗
Stack
DATA ENGINEERING
PythonApache BeamPySparkGoogle Cloud Storage
ANALYTICS / ML
BigQuery ML (ARIMA_PLUS)80% Prediction Interval
GENAI / NLP
LangChainFAISSVertex AI (Gemini)
BACKEND
FastAPIPostgreSQLUvicornAsyncIO
FRONTEND
Next.jsReactTypeScriptRecharts
DEPLOYMENT
DockerGoogle Cloud RunGCSDataflow
What I Built
An end-to-end platform that forecasts demand for 1,115 Rossmann stores using ARIMA_PLUS, delivers GenAI-powered insights, and monitors performance with real-time dashboards.
The Difference
Combines BigQuery ML forecasting, real-time pipelines, and GenAI narratives with business context retrieved for faster, data-driven retail decisions at scale.
Numbers That Matter
1,115real Rossmann stores
30days forecast horizon
33,450holdout rows (for backtest)
16.92%best local MAPE achieved
80%forecast confidence interval
Engineering Notes ✎
Real-time Data Pipeline
Apache Beam pipelines process events from multiple sources and write cleaned data to BigQuery for downstream analytics.
Forecasting Engine
BigQuery ML ARIMA_PLUS models are trained per store with 80% prediction intervals for accurate 30-day demand forecasts.
GenAI Insights
LangChain + FAISS provide relevant business context to the LLM, generating grounded, actionable narratives for stakeholders.