Semantic Anchors: An Enterprise Architecture Management Methodology for Efficient Communication with Large Language Models

This article introduces the innovative "Semantic Anchors" methodology developed by the Enterprise Architecture Management course at Bern University of Applied Sciences in Switzerland. By establishing standardized terminology, methodologies, and frameworks as reference points, it addresses the problem of understanding deviations in communication between large language models (LLMs) and enterprise architecture management (EAM), improving efficiency and accuracy. The project provides complete practical guidelines and Typst document templates, and is open-sourced on GitHub.

With the widespread application of LLMs in enterprise environments, EAM involves a large number of professional terms, methodologies, and frameworks. The lack of unified communication standards easily leads to model understanding deviations and unstable output quality. The EAM course team at Bern University of Applied Sciences in Switzerland developed the Semantic Anchors methodology to address this challenge, along with supporting document templates and CI/CD processes, providing a practical solution.

Definition of Semantic Anchors: Standardized professional terms, methodologies, and frameworks used when interacting with LLMs, serving as precise reference points to ensure consistent and accurate communication. These include standard frameworks like TOGAF, core EAM concepts, methodologies such as ADM, and best practices.

Technical Implementation:

• Using Typst instead of LaTeX for typesetting, with advantages including fast compilation, concise syntax, clear error prompts, and Git-friendliness;
• Modular document structure: main.typ (root file), template.typ (layout), chapters/ (chapters), tables/ (tables), refs.bib (references);
• CI/CD automation: GitHub Actions for continuous integration (automatic PDF compilation), version release, and artifact management.

1. Architecture Document Generation: Reference TOGAF ADM anchors in prompts to generate an outline of architecture vision deliverables that comply with standards;
2. Cross-module Knowledge Integration: Map various EAM course modules (strategic alignment, application architecture, etc.) through anchors to help understand module dependencies;
3. AI-assisted Toolchain: Combine PlantUML/Mermaid (diagrams), ChatGPT/Claude (content generation), and GitHub/Typst.app (collaboration) to form a complete workflow.

It is recommended to implement in four phases:

1. Establish Anchor List: Organize commonly used terms, methodologies, and frameworks in the organization, and categorize them with reference to project templates;
2. Formulate Usage Specifications: Clarify when to use standard terms, how to reference external frameworks, and the process for updating and maintaining anchors;
3. Integrate Prompt Engineering: Reference anchors in daily prompts and observe improvements in output quality;
4. Continuous Optimization: Expand and optimize the anchor list based on feedback to form an organization-specific knowledge base.

Semantic Anchors represent a new paradigm of AI collaboration that proactively guides LLMs to understand professional contexts. Suitable for highly specialized fields like EAM, it improves the efficiency and quality of AI assistance. This open-source project provides a theoretical framework and complete practical templates (document structure, CI/CD configuration, examples), making it an important reference resource for enterprise architecture practitioners to integrate into LLM practices.