The Cherenkov Telescope Array Observatory (CTAO) will provide incredible opportunities for the future of ground-based very-high-energy gamma-ray astronomy. To optimise its scientific output, the CTAO will have a Science Alert Generation (SAG) system, which as part of the Array Control and Acquisition (ACADA) system will perform reconstruction, data quality monitoring and scientific analysis in real-time to detect and issue candidate science alerts. As part of the continuous research and development activity for improvements of future versions of the ACADA/SAG product, this work aims at implementing machine learning enhancements for the scientific analysis. In real-time technical and observational variability, as well as performance requirements, can highly impact the overall sensitivity of the automated pipelines. We developed two prototypes of Convolutional Neural Network based models with the aim of removing any a priori knowledge requirements that standard scientific tools have. The first model is an autoencoder trained to remove background noise from counts maps of a given observation, without requiring inputs on target position, background templates or instrument response functions (IRFs). The second model is a 2-dimensional regressor that extracts hotspots for the localisation of candidate sources in the field of view, without requiring inputs on the background template or IRFs. To verify both models we use the current version of ACADA/SAG (rel1), finding that they achieve comparable results with the additional benefit of not requiring a priori knowledge.