EU AI Act Compliance Checker: An Intelligent Compliance Assessment Tool for AI Systems Based on RAG

Key Points: The EU AI Act Compliance Checker is an open-source tool developed by Crris07 and released on GitHub on June 1, 2026. Based on Retrieval-Augmented Generation (RAG) technology, it can automatically check whether an AI system description complies with the requirements of the EU Artificial Intelligence Act (EU AI Act), providing a practical assistant for AI governance. Its core value lies in combining legal provision retrieval with intelligent analysis, lowering the threshold for compliance checks, and supporting local deployment to ensure data privacy.

Background: In 2024, the EU formally adopted the world's first comprehensive law regulating AI systems, the EU Artificial Intelligence Act (EU AI Act), which uses risk-based classification (unacceptable, high, limited, minimal risk). Traditional compliance checks require legal experts to compare against the act article by article, which is time-consuming, labor-intensive, and prone to errors. This tool aims to use AI technology to automate the compliance assessment process and solve the compliance challenges faced by developers and enterprises.

Technical Architecture and Core Features:

1. Dual-layer retrieval mechanism: Combines Chroma semantic retrieval (to understand deep meaning) and BM25 keyword retrieval (for precise matching of legal terms such as Article 5 and Annex III);
2. Legal knowledge base: Obtains act texts from EUR-Lex, splits them into chunks by chapter/appendix while retaining metadata, and performs internal semantic sub-chunking;
3. Local LLM engine: By default, uses Ollama to run the Llama3.2 3B model (supports Mistral API as a fallback), enabling data privacy, offline availability, and cost control;
4. Structured report: Outputs relevant clause citations, risk levels, evidence analysis, recommended measures, as well as original text excerpts and links.

Practical Application Cases:

• Biometric systems: An AI system for real-time facial matching against a blacklist in shopping malls was assessed as potentially violating Article 5 (prohibition/strict restriction of real-time biometrics in public places) and Annex III (high-risk biometric use cases);
• Recruitment AI tools: An AI system for automatic resume screening was identified as high-risk, requiring compliance with transparency, human supervision obligations, and data governance standards.

Project Significance and Insights:

• For developers: Compliance should be a core part of product design, not an afterthought. Integrating checks can identify risks early;
• For regtech: RAG technology provides an interpretable regulatory path, and displaying original legal texts enhances transparency;
• For the open-source community: The fully open-source architecture supports local deployment, protects sensitive information, and provides a foundation for global developers to participate in building AI governance tools.

Conclusion: With the full implementation of the EU AI Act, the AI industry has entered a new compliance-driven era. This tool not only lowers the threshold for compliance but also demonstrates the possibility of AI technology feeding back into AI governance—using intelligent tools to solve regulatory challenges in the intelligent era. AI product teams targeting the EU market should pay attention to and try this open-source project.