Based on repository activity, growth velocity and community engagement.
22
Growth2/30
Activity4/25
Popularity6/25
Trust11/20
33
Stars
High
Sentiment
Votes
33
README.md
⚖️ EU AI Act Compliance Navigator
A RAG-based compliance assessment tool for the EU Artificial Intelligence Act (Regulation 2024/1689). Users describe their AI system in natural language, and the tool retrieves relevant legal provisions and generates a structured compliance report.
Why legal-structure-aware chunking? Standard RAG systems split text by
fixed token length. EU AI Act provisions have hierarchical dependencies —
Article 6's classification rules reference Annex III's high-risk list,
which in turn triggers obligations spread across Articles 9–14. Fixed-length
splitting severs these connections. The chunker in
respects Chapter, Section, Article, and Annex boundaries, preserving the
legal logic each chunk carries.
src/legal_chunks.py
Why Self-Query routing instead of pure semantic search? Annex III
(high-risk AI system categories) and Annex IV (technical documentation
requirements) are semantically similar in embedding space — same legal
register, similar structure — but their legal consequences are mutually
exclusive. Semantic-only retrieval conflates them. The regex router detects
explicit legal references and applies metadata filters before vector search,
eliminating this class of error for direct citation queries.
Key Features
Chat Interface — Conversational UI with user messages right-aligned
6 Languages — English, French, German, Spanish, Simplified Chinese, Traditional Chinese
Self-Query Routing — Mentions of “Article 6”, “Annex III”, etc. trigger metadata-filtered search
Dynamic Token Budget — Auto-adjusts context window (6K–10K tokens) based on query complexity
Quick Query Buttons — Pre-built scenarios for common compliance questions
Download Answers — Export individual answers or full chat history as Markdown
Regenerate in New Language — Switch language and regenerate any previous answer
Screenshots
External Showcase
To present the project in a public evaluation setting, this prototype was submitted to Hack Trek 2026 on Devpost.
This submission reflects the project’s transition from a private prototype to a publicly demonstrable legal-tech tool for EU AI Act compliance analysis.
Run python run_pipeline_chroma.py to generate embeddings
Export updated vector store:
python -c "
import json, chromadb
db = chromadb.PersistentClient(path='./chroma_db')
col = db.get_collection('eu_ai_act')
data = col.get(include=['documents','metadatas','embeddings'])
export = []
for i in range(len(data['ids'])):
emb = data['embeddings'][i]
if hasattr(emb, 'tolist'): emb = emb.tolist()
export.append({'id':data['ids'][i],'document':data['documents'][i],'metadata':data['metadatas'][i],'embedding':[round(x,5) for x in emb]})
with open('vector_store.json','w') as f:
json.dump(export, f, separators=(',',':'))
print(f'Exported {len(export)} records')
"
Push updated vector_store.json to GitHub
Example Questions
“What does Article 6 say about high-risk AI classification?”
“An AI system that screens job applicants’ CVs and ranks candidates.”
“What obligations do providers of high-risk AI systems have?”
“What AI systems are listed in Annex III?”
“What transparency requirements apply to AI systems?”
“What AI practices are prohibited under the EU AI Act?”
“Does my biometric identification system fall under Annex III?”
“What does Article 9 require for risk management?”
Legal Documents Coverage
Primary Legislation (EU AI Act)
| Document | Notes |
|---|---|
| EU AI Act — Full Text (Regulation 2024/1689) | Articles 1–113 |
| EU AI Act — Recitals 1–180 | Indexed separately for precision retrieval |
| EU AI Act — Annexes I–XIII | Indexed separately; Annex III (high-risk list) and Annex IV (technical documentation) treated as distinct retrieval targets |
Cross-Regulatory Framework
| Document | Relevance |
|---|---|
| GDPR (Regulation 2016/679) | Data governance obligations intersecting high-risk AI systems |
| Cyber Resilience Act (Regulation 2024/2847) | Cybersecurity requirements cross-referenced in AI Act |
| Product Liability Directive (2024/2853) | Liability framework for AI-embedded products |
| AI Liability Directive — Proposal (COM/2022/496) | Proposed civil liability regime for AI systems |
| Machinery Regulation (2023/1230) | Safety requirements for AI integrated into machinery |
Official Guidance
| Document | Source |
|---|---|
| EU AI Act FAQ / Implementation Guidelines | European Commission |
| Codes of Practice for General-Purpose AI | EU AI Office |
| AI Act Compliance Checker Flowchart v1.0 (Updated July 2025) | Future of Life Institute — decision tree mapping entity types, risk classification, and obligation triggers across Articles 5, 6, 25, 50, 51 |
Academic Sources
| Document | Source |
|---|---|
| "Truly Risk-Based Regulation of Artificial Intelligence: How to Implement the EU's AI Act" (Working Paper No. 101) | Martin Ebers — Stanford–Vienna Transatlantic Technology Law Forum |
| "Interplay between the AI Act and the EU Digital Legislative Framework" | European Parliament, ITRE Committee Study |
| "Systemic Data Bias in Real-World AI Systems: Technical Failures, Legal Gaps, and the Limits of the EU AI Act" | Falelakis, Dimara & Anagnostopoulos — MDPI Information |
Total: 14 documents → 3,230 indexed passages
Background
The EU AI Act (Regulation 2024/1689) enters full enforcement on 2 August 2026.
With 113 articles, 180 recitals, and 13 annexes, it presents a concrete
navigability problem: organisations need to determine whether their AI system
is high-risk, what obligations apply, and where exactly those obligations
are located across a complex legislative structure.
This tool is an attempt to make that structure computationally navigable —
a question at the intersection of regulatory law and retrieval system design.
Built by a law student at Jilin University (Changchun, China).
Disclaimer
This tool provides preliminary guidance only and does not constitute legal advice. Always consult qualified legal counsel for compliance decisions.
Ecosystem Role
Standard MoltPulse indexed agent.
Embed Badge
Show off your Pulse Score in your GitHub README to build trust and rank higher.
