The Hydr.IA project involves the SYNAPSE Company and the HydroSciences Montpellier Laboratory. It is based on a specific observation: despite the growing development of AI-related solutions in many complex fields, and the demonstration of their relevance to a number of hydro(geo)logy issues, there is a deficit in the transfer of these technologies to end-users. It would appear that two elements are lacking to ensure a more massive transfer of these solutions to the market. The first is the difficulty for research to be directly confronted with real operational conditions. The second is the limited confidence users have in solutions that are often perceived as opaque, and whose operational implementation is rare, leading to a vicious circle that this project aims to break. The program of work envisaged by this project will meet the following objectives: 1) Definition of the range of real-time forecasting services, corresponding to the need for impact reduction and in line with market acceptance, 2) Creation of a demonstration platform for these services using digital twins to demonstrate the efficiency of AI to end-users, 3) Development of methodologies for the design of AI models adaptable to the range of services defined above, 4) Extend the forecasting horizon, currently limited to the response time of the hydrosystem, by coupling AI modeling solutions with meteorological forecasts that can be post-processed. The city of Nîmes, in the South of France, is the first study site chosen. Mediterranean climate combined with small karst catchments can lead to complex flash flood events (concentration time less than 1/2 hour) resulting in large damages. The city shows the particularity to be crossed by different urban rivers. The city also acquired an alert system to support crisis management and better predict floods. This system, named ESPADA, uses a hydrological model including karst components, high-resolution precipitation and hydrometry measurements on the entire catchment, in real time. By reducing costs from field studies and physical model conception, the Hydr.IA approach would allow the application of low cost and effective tools for floods risk management. The poster will present the Hydr.IA project and the first results in comparing and combining both IA and conceptual approaches for Nimes floods forecasting.