This study presents a comprehensive framework for applying Artificial Intelligence (AI) to create graph-based dimensional representations within abstract spaces for visualizing and managing public projects. It addresses the critical challenge of automating project classification by transforming it into an AI-driven knowledge representation problem, converting unstructured textual project descriptions into structured, spatial formats. Key findings reveal that embedding-based models, specifically pre-trained Polish-language HerBERT model, offer the most viable approach for initial implementation, effectively balancing performance with practical constraints while capturing semantic relationships between project descriptions and strategic goals. This approach significantly overcomes the limitations of traditional manual, keyword-based classification systems. The research establishes a robust conceptual framework supported by a comparative analysis of NLP and machine learning methods, providing clear implementation guidelines including dataset curation from the Silesian Voivodeship, accuracy evaluation using precision, recall, and F1-score metrics, and iterative refinement processes for domain-specific terminology. The proposed solution extends beyond simple project monitoring to form the foundation of a comprehensive knowledge management system, where abstract spatial maps serve as dynamic AI-driven knowledge representations. This system provides decision-makers with objective, data-driven feedback loops, enabling more effective strategic alignment reasoning. The work establishes both theoretical foundations and practical implementation pathways for transforming unstructured public administration data into structured, visualizable insights that enhance efficiency, objectivity, and consistency in project monitoring and strategic reporting.