Real-Time Demand Forecasting System

Project Type: Real-Time Pipeline | Forecasting | Time Series | Dashboard
Tools Used: Python, Pandas, Prophet, PostgreSQL, Supabase, Kafka, Docker, Streamlit, Cron
Dataset: Simulated Order Data Streamed in Real-Time

Link to Live Dashboard

Objective

Build a full-stack real-time demand forecasting system capable of simulating orders, aggregating and storing metrics, generating daily forecasts, and visualizing results in an interactive, cloud-deployed dashboard.

Key Features

• Real-Time Event Streaming:
  • Used Kafka to simulate a continuous stream of customer orders with custom payloads (region, category, user segment, price).
  • Kafka consumer stores incoming events into a PostgreSQL database.
• Data Aggregation and Forecasting:
  • Daily metrics (total revenue, total orders, average order value) aggregated and inserted into a dedicated daily_metrics table.
  • Forecasts generated using Prophet and stored in forecast_metrics with uncertainty intervals.

• Simulated Promotions and Segments:

  • Injected realistic behavior by simulating weekends, holidays, and promotional days.
  • Orders contain metadata like hour, category, and user segment to allow detailed filtering.

• Automation & Scheduling:

  • Full pipeline automated with a cron job to run daily at 10:30 AM.
  • Includes a shell script to activate the virtual environment and sequentially trigger each pipeline step.
• Streamlit Dashboard:
  • Filter by region, category, user segment, hour of day, and lookback window.
  • Visualize historical and forecasted metrics with clear promo overlays.
  • Download filtered CSVs and explore tabular forecast data.

Deployment Highlights

• Streamlit Cloud used for dashboard deployment.
• Supabase hosted PostgreSQL used for fast and secure data storage.
• Dockerized Kafka/Zookeeper/Postgres stack for local simulation.

🔗 View Blog Post: Link
🔗 View Code on GitHub: GitHub Link