Ms Somin Lee (p), Mr Donghyun Kim

Ex-ante evaluation of large-scale public investment projects is a critical institutional mechanism for assessing policy relevance, economic viability, and fiscal accountability prior to implementation. However, prolonged review periods and complex administrative procedures frequently delay decision-making, increasing social costs and weakening the timely provision of essential infrastructure and public services. Although recent advances in large language models (LLMs) have opened new possibilities for automating knowledge-intensive policy tasks, conventional LLM applications encounter significant limitations when processing extensive regulatory documents and structured evaluation frameworks. In particular, context rot and hallucination can undermine analytical reliability and institutional consistency in high-stakes public decision environments.
To address these challenges, this study proposes a Context Engineering–based AI agent framework for automating structured public investment evaluation. The framework organizes the assessment process into four coordinated stages. First, during the knowledge ingestion phase, institutional guidelines, evaluation criteria, and standardized cost references are retrieved through a Retrieval-Augmented Generation (RAG) mechanism to ensure traceability and regulatory alignment. Second, in a multi-agent collaborative analysis stage, specialized agents responsible for demand forecasting, cost estimation, and policy assessment exchange data through a Shared Unified History, maintaining analytical coherence across modules. Third, Recursive Summarization selectively transmits essential information between stages to prevent context degradation during long-document generation. Finally, Memory Consolidation integrates intermediate outputs into a logically consistent final evaluation report.
Empirical testing using real-world public project proposals demonstrates that the proposed system generates evaluation documents with high structural integrity and alignment with institutional standards. Furthermore, the framework produces confidence intervals for key performance indicators, such as benefit–cost ratios, enabling pre-submission scenario simulations and early identification of potential vulnerabilities.
The findings indicate that context-engineered AI agents can substantially reduce administrative lead times while strengthening reliability, transparency, and traceability in public investment decision-making. More broadly, this research advances the methodological integration of LLM-based systems into institutional policy evaluation, positioning context-aware AI as a scalable governance infrastructure for enhancing efficiency and consistency in the public sector.

Ms Gülsüm Yavuz (p), Professor Tüzin Baycan

Natural disasters constitute one of the most critical challenges confronting contemporary societies due to their direct impacts on human life, the spatial destruction they cause, and their long-term economic consequences. Urban areas, in particular, face disproportionately severe outcomes as a result of concentrated populations, dense infrastructure systems, and intensified economic activities. Processes such as rapid and uncontrolled urbanization, environmental degradation, and climate change have significantly increased both the frequency and severity of disasters, transforming disaster risks into more widespread, complex, and interdependent phenomena compared to previous periods. This transformation necessitates addressing disasters not merely as extraordinary events requiring post-event response, but as dynamic risk processes that must be systematically analyzed, managed, and mitigated in advance.
Within the pre-disaster phase of disaster management, the identification of risks, the spatial analysis of risk patterns, and the development of risk reduction-oriented policies have become increasingly critical. However, the growing diversity of risks and the complexity of spatial risk configurations have made the limitations of conventional analytical approaches more evident. The processing of large-scale spatial datasets, the modeling of multivariate relationships, and the enhancement of predictive capacity require more flexible and advanced analytical tools. In this context, despite certain methodological limitations, artificial intelligence (AI) technologies offer substantial potential for risk prediction, the analysis of complex interrelations among risk factors, and the production of high-resolution risk maps. Indeed, recent years have witnessed a rapid increase in the use of AI-based approaches in urban disaster risk analysis. The expanding body of literature calls for a comprehensive assessment that reveals methodological trends and research patterns within this field.
Accordingly, this study systematically reviews research focusing on the analysis of spatial disaster risks in urban areas using artificial intelligence algorithms. Studies selected according to predefined criteria are thematically analyzed in terms of risk categories, algorithm preferences, data types, spatial scaling approaches, and validation methods. Based on this analysis, a multidimensional taxonomy is developed for the AI-based assessment of urban spatial disaster risks. The study aims to provide a guiding methodological reference for researchers and to offer an analytical perspective for decision-makers in the fields of planning and disaster management.

Prof. Elisabeth Baier (p)

In recent years, the rapid diffusion of artificial intelligence (AI) has become a key driver of economic and societal change. At the same time, economic geography has witnessed a growing interest in imaginaries, narratives and visions as drivers of regional development (Benner, 2024). Narratives are socially shared sense-making stories connecting events, actors and actions, thereby shaping which regional development paths appear plausible (Sotarauta & Grillitsch, 2023).
In addition to these developments, little conceptual or empirical research has systematically engaged with alternative regional futures and future-making practices (Gong, 2024). Moreover, despite the rapid diffusion of AI, the enabling and constraining effects of generative AI remain underexplored in the context of regional development, particularly in narrative-oriented approaches.
This contribution addresses this gap through an exploratory application to two structurally challenged regions in Germany, Lusatia and the Ruhr Area. Generative AI is conceptualized as a methodological instrument for narrative diagnosis and narrative design. Drawing on large language models (LLMs) such as GPT-5.2, the paper develops and illustrates a structured framework for AI-assisted narrative analysis and recombination for regional development and the exploration of alternative regional futures, while critically reflecting on its opportunities and limits.
It first discusses the role of narratives and generative AI in shaping regional trajectories, then examines their application potential for narrative mapping, narrative recombination and counterfactual narrative generation. In both regions, LLM-assisted narrative mapping is used to identify dominant development frames and marginalized storylines embedded in regional strategy and policy documents. These narrative configurations are then algorithmically recombined to generate counterfactual, adjacent possible futures, which are critically assessed with regard to their plausibility and transformative potential.
The empirical application enables a critical reflection on how generative AI can function as a reflexive “sparring partner” expanding the imaginative space of regional actors while leaving normative evaluation and strategic choice in human hands. Overall, this contribution advances the understanding of AI-assisted narrative co-creation in regional development contexts and proposes a structured methodological approach for broadening the imaginative and methodological horizons of regional research and future-making policy practice.
References
Benner, M. (2024). An ideational turn in economic geography? Progress in Economic Geography, 2(1), 100014.
Gong, H. (2024). Futures should matter (more): Toward a forward-looking perspective in economic geography. Progress in Human Geography, 48(3), 292–315. h
Sotarauta, M., & Grillitsch, M. (2023). Path tracing in the study of agency and structures: Methodological considerations. Progress in Human Geography, 47(1), 85–102.