Wasail: Demand Forecasting System

I developed Wasail, a demand forecasting system designed for grocery stores in Pakistan to optimize inventory and minimize both overstocking and understocking. Inspired by Y Combinator startups like Shelf Engine and Guac, the project was built after extensive research and experimentation, covering data cleaning and preprocessing, exploratory data analysis, feature engineering, model training, hyperparameter tuning, model evaluation, and deployment. The core goal of Wasail is to empower grocery stores with smarter data-driven decisions that reduce waste and improve efficiency.

For forecasting, I used Darts, an advanced time series forecasting framework, to implement a wide range of models including statistical, regression, and deep learning approaches. The statistical models included ARIMA, Exponential Smoothing, and Prophet, while the regression models included Linear Regression, Random Forest, LightGBM, XGBoost, and CatBoost. On the deep learning side, I implemented state-of-the-art architectures such as N-HiTS, TCN, Transformer, D-Linear, N-Linear, TiDE, and TSMixer. These models were trained on diverse datasets incorporating historical sales data, oil prices, holidays, promotions, and store locations. Model selection was guided by the latest trends in time series forecasting and tailored for practical application in the retail environment. Darts also provided powerful utilities such as trend and seasonality detection, grid search, backtesting, historical forecasts, and forecast accuracy metrics, all of which strengthened the system’s reliability.

To bring this system into real-world use, I integrated AI into a grocery store mobile application using a microservice architecture, with a Flask API deployed on DigitalOcean. The API serves as the backbone for communication between the mobile app and the AI models, enabling seamless real-time data processing and insightful analytics. By leveraging advanced models such as LightGBM, Prophet, and XGBoost in production, the app delivers accurate demand forecasting and inventory management capabilities. This integration ensures that store owners and managers receive actionable insights directly in their hands, ultimately providing a smarter, more efficient, and user-friendly experience.