The main outputs from the project in this period have been 19 refereed publications. These range from foundational works and preparatory studies for exploiting the upcoming LSST data, to first high-impact results. (1) The project has made major strides in developing a new and powerful stellar population synthesis (SPS)-based machine-learning model for accurate redshift distributions and galaxy population properties. We have calibrated the model with the state-of-the-art COSMOS2020 26-band galaxy catalogue, and are now in the process of applying this approach to deriving weak lensing cosmological constraints using the KiDS-Viking catalogue. (2) We improved the state-of-the-art constraints on light dark matter by developing a robust, fast and accurate emulator of Lyman alpha forest data based on hydrodynamical simulations. The emulator methodology we developed has been widely adopted in the literature for diverse science goals. (3) We have published a fast and realistic population model self- consistently linking optical signals from gravitational wave sources, and used this to inform LSST observing strategy; we have made critical contributions in understanding the main levers in the LSST observing strategy for classifying long-lived transients such as supernovae, which is key to eventual scientific exploitation of the LSST data. (4) We have used a novel explainable artificial intelligence method to improve the understanding of the emergent/universal properties of dark matter halos, and the connection between their emergent properties and their evolution histories. These results are key steps in connecting the physics of dark matter to the observed properties of non-linear structures such as galaxies and clusters of galaxies.