Ralph Spencer Steenblik
Architectural practice is a knowledge-intensive discipline that generates, consumes, and continuously revises information across radically different registers: precedent images, code requirements, material specifications, structural logic, client narratives, and pedagogical history. Yet the dominant tools for managing this knowledge (email threads, cloud folders, keyword search) treat information as a collection of discrete documents rather than as a network of interdependent propositions. The result is systemic fragmentation: teams carry institutional memory in individual heads, knowledge does not accumulate across projects, and the relational logic that defines design problems is routinely lost.
This paper presents a working prototype (the phi GraphRAG) as a case study in what we term Retrieval-Augmented Design (RAD): the application of graph-native knowledge representation and retrieval to architectural practice. Built in collaboration with computer science graduate students at Indiana University, the phi GraphRAG ingests project documents, extracts entities and relationships, and represents them as a queryable knowledge graph. When a designer queries the graph, the system returns not isolated documents but relational context: the web of constraints, precedents, and decisions that surround any given question.
We argue that GraphRAG is not merely a productivity tool but a systemic design artefact. It makes the knowledge architecture of practice visible, navigable, and improvable. Drawing on systemic design theory, we position the knowledge graph as an information metabolism layer that connects design memory (what the firm knows) to design action (what the team is deciding now). Implications for collaborative practice, pedagogy, and the “systems that fit systems” research agenda are discussed.
Keywords: GraphRAG, knowledge graph, retrieval-augmented generation, architectural practice, systemic design, collective memory, artificial intelligence