From Analysis to Application: Employing AI to Enhance User Experience at ESA

In the current era, Artificial Intelligence (AI) has gained remarkable traction, revolutionizing tasks such as translation, summarization, and text generation. This master's thesis delves into the application of AI technologies to enhance User Experience (UX) in European Space Agency (ESA) applications, specifically those managed by EOP-GES (Ground Segment). The thesis comprises two distinct projects and a demonstrator designed to showcase the practical implications of these projects in real-world scenarios. The first project, EO Recommender, uses Earth Online visitors' navigation history to predict the most likely pages they will visit. The activity includes collecting and curating an ad-hoc EO Clicks dataset capturing user activity on the website. We then employ state-of-the-art recommendation systems such as Gru4Rec and SLiRec. We provide extensive benchmarking and hyperparameter tuning and assessment. The EO Recommender's development yields valuable insights for ESA and demonstrates the potential for enhancing user engagement through this technology. The second project, the EOP-GES Assistant, introduces an offline virtual assistant powered by Large Language Models (LLMs), which draw knowledge from private document collections, ensuring security. We adopt an open experimentation approach to enable the testing of such a system with real data. Specifically, we collect prompts and corresponding related information and documents from ESA staff. We then enhance and generate new prompts to enlarge the experimental dataset. Finally, we benchmark multiple LLMs and analyze their results in terms of quality, time, and robustness. Benchmarking of models provided insights into resource requirements and trade-offs for implementing similar systems. Among the tested LLMs, LLaMA V2 models reveal their remarkable capacity to extract nuanced information from documents of diverse genres. We include both projects in a web application to effectively showcase their usability and practical implications of our projects. Our analysis and insights offer a promising starting point for further research and implementation endeavors at EOP-GES.