Artificial intelligence is rapidly improving weather prediction, but translating global forecast information into local, decision-relevant guidance remains a major challenge. This seminar presents a multi-timescale research framework spanning radar nowcasting, kilometer-scale short-range prediction, and subseasonal forecasting. I will introduce AirCast-SR, a generative downscaling system that transforms global AI weather forecasts into hourly fields at approximately 1 km resolution while preserving fine-scale structures in precipitation, temperature, winds, and solar radiation. I will also discuss radar- and satellite-informed deep learning approaches for 0–3-hour precipitation nowcasting and AI-enabled methods for extracting regional signals from multi-model subseasonal forecasts. Examples from the United States and South Asia will illustrate potential applications to floods, extreme heat, renewable energy, and disaster preparedness. The presentation will also examine uncertainty, physical consistency, observational limitations, regional transferability, and operational deployment; key challenges in converting advances in AI weather prediction into reliable and actionable environmental information.