Dye-sensitized solar cells (DSCs) are a photovoltaic technology based around light-harvesting dye molecules bound to thin semiconductor films of high surface area. Many of the highest-performing DSCs to date incorporate multiple dyes that harvest light from different regions of the solar spectrum in a complementary manner – these are known as cosensitized DSCs. However, finding dyes that are well-suited for cosensitization is a long and costly experimental process when carried out through trial and error in a laboratory. To help direct experimentalists towards promising candidates, the main project of this thesis harnesses ideas from data-driven materials discovery to develop an entirely computational pipeline that predicts boosts in performance of dye pairs when cosensitized. It does this by identifying partner dyes that show the most complementary absorption characteristics to sets of well-known or high-performing starting dyes, systematically sifting candidates from a large database of optically active compounds. It then uses density functional theory (DFT) simulations to compute key structural, electronic and optical properties of the selected pairs of dyes, which are used as inputs to models that predict short-circuit current density (JSC) and open-circuit voltage (VOC), two key device performance parameters. The predictive models for JSC and VOC of singly-sensitized devices are developed further from existing models used in previous works, and are also expanded to the cosensitized case for the first time. 11 starting dyes were passed through the pipeline (six organic and five organometallic), leading to 22 dyes in total being modelled at the DFT level as 11 pairs. The accuracy of predicted JSC and VOC for single sensitizers was tested against existing experimental references. Notably, half of the JSC predictions were within 20% error or less of experimental values whilst others had greater discrepancies, the sources of which are discussed in detail. These results are significant given the choice of structurally dissimilar dyes here – this accuracy is on par with previous computational studies that focussed only on sets of structurally analogous dyes. From the predictions of cosensitized devices containing the complementary dye pairs, two standout cells were those containing **SQ2**+**LD2** dyes and **YD2**+**VKXB** dyes, which gave +13% and +12% boosts to JSC relative to their singly-sensitized counterparts, respectively. A secondary computational project was also carried out in collaboration with previous experiments of DSC dye monolayer growth over time. Whilst complete dye monolayers have been studied extensively, their behaviour as they grow is less well understood, despite its importance for DSC fabrication. X-ray reflectometry (XRR) had been used by a collaborator to investigate monolayer thicknesses and densities as they grow under different conditions in the DSC fabrication process. This author trained a neural network to perform rapid, deterministic fitting of 360 experimental reflectivity curves in high-throughput fashion. The DSC dye layer parameters predicted by this machine-learning model were compared to those from a human-assisted fit with standard software (such fitting being orders of magnitude slower to carry out). The neural network predictions had high accuracy for instances where monolayers adhered to the assumptions of the Parratt model used to fit reflectivity curves, but poorer accuracy during periods of faster change in thickness, suggesting dynamic behaviour of dye ensembles that warrants further investigation. Thus, the neural network acted as a supporting tool to identify where to focus further experimental DSC investigation, which is the overarching theme connecting the two projects of this thesis. Chapter 1 provides a literature review of DSC function, the structure-property relationships of their component materials, and pre-existing computational methods that predict DSC performance. Chapter 2 provides a technical background to the density-functional theory (DFT) methods used throughout much of this work. Chapter 3 presents the design-to-device pipeline methodology developed in this work. Chapter 4 displays and discusses the results of this pipeline as applied to six well-known or high-performing organic dyes and their six complementary partner dyes identified. Chapter 5 similarly presents results for five ruthenium-based dyes and their cognate organic partner dyes that were identified by the pipeline. Chapter 6 provides a background to XRR and neural networks, before presenting the training of neural network and evaluating its performance in reproducing fitted layer parameters from the experimental XRR data described above. Chapter 7 discusses the conclusions of this work and how further research may be enabled.