Interfacing reference severe accident (SA) codes with commercial nuclear simulators can help spread SA knowledge to a larger audience and build more accurate plant models. However, this requires improving the performance of calculation codes. The Horizon Euratom ASSAS project (Artificial intelligence for the Simulation of Severe AccidentS) addresses this issue and evaluates, among other strategies, the possible added-value of machine-learning (ML) to speed-up calculations. Efficient programming will be applied to ASTEC (Accident Source Term Evaluation Code, developed by IRSN), with negligible impact on accuracy. This might be sufficient to achieve a real time execution of ASTEC if combined with simplifications of some physical models and of the nodalization used in the input decks. Nevertheless, such modifications are hard to implement, and high acceleration factors are out of reach for complex legacy multiphysics codes like ASTEC. ML could be a game-changer to reach significantly higher acceleration factors. The first step is the definition of the parts of the SA code that can be replaced by a data-driven surrogate model. Global models, replacing the code integrally, can be more efficient and do not require access to the source code. They will be tested on ASTEC for the ex-vessel phase of the accident. On the other hand, they are expected to be data-greedy since they need to capture a large diversity of physical phenomena. Therefore, ASSAS also explores hybrid approaches, for which surrogate models will replace only selected functions of ASTEC and will be interfaced with other modules. The specifications of the basic-principles simulator to be developed during ASSAS must be selected at the beginning of the project, to ensure the consistency between the plant model and the training data for ML models. Exercise scenarios have been chosen to determine the systems to be modelled, their scope and the data to be displayed on the Human-Machine Interface. The present paper aims at providing a description of the ASSAS project and its goals. The adopted strategy and the related challenges will be discussed.