AI Regulatory Decision Support Systems

Explainable AI that helps public-sector professionals interpret regulations, analyse complex documents and make transparent, evidence-based decisions while preserving human oversight.

Comment

Inspiration

Public-sector and regulatory professionals work with complex legislation, technical standards, administrative procedures and large volumes of documentation. Much of this work is still performed manually, even though it requires repeated cross-checking, interpretation and evidence-based reasoning.

This project was inspired by direct professional experience in public safety, fire safety compliance and public procurement. The objective is to explore how Explainable AI can support professionals in demanding regulatory environments without replacing human judgement or responsibility.

What it does

AI Regulatory Decision Support Systems is a human-centred framework for applying Explainable Artificial Intelligence to regulatory compliance.

The proposed system combines:

  • Large Language Models
  • structured regulatory knowledge
  • document intelligence
  • transparent reasoning
  • human review

It analyses complex documentation, identifies applicable legal and technical requirements, detects inconsistencies and produces explainable findings supported by evidence and regulatory references.

Two representative use cases are presented:

  1. Fire Safety Compliance Assistant
  2. Public Procurement Compliance Assistant

How it was built

The project was developed as a research and prototype concept using OpenAI models and Codex-assisted workflows.

The architecture includes:

  • Document Intelligence Layer
  • Regulatory Knowledge Layer
  • Regulatory Reasoning Engine
  • Compliance Verification Engine
  • Explainability Layer
  • Human Review Layer

The system is designed as decision support, not autonomous decision-making. Every recommendation remains traceable, reviewable and subject to professional validation.

Challenges

The main challenge was translating complex regulatory workflows into a reusable AI architecture while preserving explainability, legal traceability and human accountability.

Another challenge was ensuring that the system does not simply retrieve information, but reconstructs the reasoning process used by experienced professionals when evaluating compliance.

Accomplishments

The project produced a complete research package including:

  • a detailed Research Proposal
  • an Executive Summary
  • a Technical Appendix
  • a Demonstration Concept
  • a Presentation Deck
  • prototype workflows and interface concepts
  • a real-world fire safety compliance analysis based on legislation, technical standards and building plans

The project demonstrates how the same architecture can be adapted across multiple regulatory domains.

What I learned

The most valuable role of AI in regulated environments is not autonomous decision-making, but transparent collaboration with qualified professionals.

Trust depends on the ability to explain each finding, identify its evidence and show the reasoning that produced it.

What's next

The next phase is to develop a functional prototype, beginning with the Fire Safety Compliance Assistant.

Future work will include:

  • structured regulatory knowledge bases
  • retrieval-augmented generation
  • multimodal document and drawing analysis
  • explainability and confidence indicators
  • professional user evaluation
  • expansion to additional public-sector domains

The long-term vision is a reusable platform for trustworthy AI-assisted regulatory decision support in public administration.

Independence statement

This project has been prepared exclusively in a personal capacity as an independent research initiative. It does not represent the views, policies or official position of the Hellenic Fire Service or any other public authority.

Built With

  • administration
  • ai
  • codex
  • compliance
  • document
  • explainable
  • fire
  • gpt-5
  • human-in-the-loop
  • intelligence
  • language
  • large
  • models
  • multimodal
  • natural-language-processing
  • openai
  • public
  • rag
  • regulatory
  • safety
Share this project:

Updates