I have three years of daily historical demand data for a natural-gas network and need an AI/ML model that can accurately forecast short- to mid-term demand. Scope • Clean and engineer the time-series data, handling gaps, outliers, and seasonality. • Develop and compare at least two forecasting approaches (e.g., SARIMAX, Prophet, LSTM, or gradient-boosted trees) and select the best performer. • Tune hyper-parameters, apply cross-validation, and document accuracy with MAE, RMSE, and MAPE. • Package the final model as a reproducible Python notebook or script with clear comments and a brief README explaining input format and how to retrain on new data. Nice-to-have While only historical demand data are provided right now, please structure the pipeline so environmental or economic variables can be added later without major refactoring. Deliverables 1. Cleaned dataset and feature-engineering code. 2. Two or more candidate models with evaluation results and comparison plots. 3. Final chosen model saved (pickle/joblib) plus inference script. 4. README summarizing methodology, assumptions, and how to run or extend the model. Tools Python 3.x with common libraries such as pandas, scikit-learn, statsmodels, TensorFlow/Keras or PyTorch—use what best fits the winning approach. I’m aiming for a functional prototype that is ready for internal testing and easy to enhance in future iterations.