Accurate prediction of daily solar irradiance is critical for smart grid stability, energy storage planning, and photovoltaic system optimization. Existing forecasting models are predominantly trained on single geographic regions, limiting their applicability across India's climatically diverse landscape. This paper presents S.U.R.A.J. (Solar Utility & Radiance Analytical Judgment), a machine-learning-based solar energy forecasting system validated across six geographically and climatically distinct Indian cities: Jabalpur, Bhopal, Delhi, Mumbai, Jaipur, and Ladakh — representing tropical, semi-arid, arid, coastal, and cold-arid climate zones. The system employs an 11-step reproducible data pipeline utilizing five years (2019–2023) of daily meteorological observations acquired from the NASA POWER satellite API, comprising 1,826 data points per city. A Random Forest Regressor is independently trained for each city using 17 engineered features, including cyclical temporal encodings, autoregressive lag variables (Solar_Lag_1, Solar_Lag_7), and rolling statistical features (Solar_Roll7, Solar_Roll30), applied to an 80/20 chronological train-test split. City-specific models achieve R² scores ranging from 0.830 (Ladakh, cold-arid) to 0.891 (Mumbai, coastal), with Mean Absolute Errors of 0.337–0.524 MJ/m², substantially outperforming the naive persistence baseline across all locations. Feature importance analysis consistently identifies cloud fraction and previous-day irradiance as the dominant predictors of daily solar output. The trained models are deployed through a Flask REST API connected to an interactive web dashboard, enabling real-time solar irradiance simulation via adjustable environmental controls. Results confirm that climate-specific ensemble modeling with engineered temporal features produces reliable, interpretable, and operationally deployable solar forecasts for diverse Indian cities.