The abstraction and formalization of architectural design thinking into structured knowledge representations remains a critical gap in computational design theory. This paper proposes a conceptual framework for translating tacit design knowledge from exemplary architectural projects and expert design processes into machine-interpretable semantic networks, aiming to bridge the epistemological divide between architectural practice and large language models (LLMs). By synthesizing principles from design cognition theory and semantic graph formalisms, the framework establishes a Design Thinking Semantic Framework (DTSF) that constructs relational graph structures that map architectural lexicons to LLM-compatible semantic units, preserving the nuanced interplay between creative intuition and disciplinary rigor inherent to architectural problem-solving. Through ontological alignment mechanisms, the framework enables the transfer of domain-specific design intentionality—such as the dialectic between form and function or the negotiation of heritage constraints—into generative AI systems. This theoretical advancement provides a structured methodology for embedding architectural epistemology into AI models, fostering cross-disciplinary dialogue while addressing the limitations of current LLM applications in architecture. The work lays foundational principles for future AI-augmented design tools that respect architectural complexity without reducing it to combinatorial optimization.