Build & Deploy a Machine Learning App | DemandLens MVP 1 Walkthrough

In this video, I walk through how I built DemandLens MVP 1 — taking a traditional machine learning forecasting model from a notebook to a deployed web application. This is a high-level walkthrough of the key steps, technologies, and engineering decisions behind the project. Rather than explaining every line of code, we’ll look at how the different components fit together to create a working end-to-end application. What we’ll cover The problem: Forecasting daily retail sales using historical data from the Kaggle Store Sales — Time Series Forecasting competition. Machine learning: Exploring the dataset, engineering time-series features, comparing model approaches, and using cluster-based LightGBM models. Recursive forecasting: Understanding how predictions for earlier dates help generate forecasts for later dates. Backend development: Using Python and FastAPI to expose the forecasting model through an API, with Pydantic for request validation. Frontend development: Building an interactive dashboard with Next.js, React, TypeScript, and Recharts. Explore DemandLens 🌐 Live application: https://demandlens-api.netlify.app/ 💻 GitHub repository: https://github.com/umeaimanMerchant/d... ⚙️ API documentation: https://demandlens-api.onrender.com/docs 📊 Dataset: https://www.kaggle.com/competitions/s... What comes next? MVP 1 establishes the forecasting foundation: a trained machine learning model, a backend API, an interactive frontend, and a deployed application. The next phase of the DemandLens journey is to build a deterministic what-if scenario planner, allowing users to explore business assumptions and compare forecast outcomes. This is part of my journey to become a stronger Applied AI Engineer by building real applications, learning the engineering behind them, and sharing the process along the way. Subscribe to ThinkWise with Iman to follow the journey. Learn. Think. Understand.