Elioz's Team
End-to-end pipeline that converts government regulation documents (PDF/HTML) into working compliance software.
Core trick
Parses regulation articles into clauses with deterministic IDs, then has an LLM extract requirements and validate them with a second-model judge gate.
Description
End-to-end pipeline that converts government regulation documents (PDF/HTML) into working compliance software. Powered by Anthropic Claude Opus 4.6 as the AI backbone for quality assurance, deep analysis, and intelligent compliance evaluation.
README excerpt
<p align="center">
<img src="image/RegulationCoder-Logo.png" alt="RegulationCoder Logo" width="600">
</p>
# RegulationCoder
End-to-end pipeline that converts government regulation documents (PDF/HTML) into working compliance software. Powered by **Anthropic Claude Opus 4.6** as the AI backbone for quality assurance, deep analysis, and intelligent compliance evaluation.
## What it Does
1. **Ingests** regulation documents (EU AI Act PDF)
2. **Parses** articles into structured clauses with deterministic IDs
3. **Extracts** requirements using Claude Sonnet 4.5 (with Judge Gate A validation by Claude Opus 4.6)
4. **Formalizes** requirements into machine-checkable rules (with Judge Gate B validation)
5. **Evaluates** AI system profiles against 53 compliance rules
6. **Analyzes** compliance posture with Claude Opus 4.6 deep AI analysis
7. **Exports** reports (JSON/HTML) with full audit trail
## Demo: EU AI Act Compliance
Evaluates a fictional recruitment AI system ("TalentScreen AI") against EU AI Act Articles 9-15 (high-risk AI requirements).
**Expected result**: Partial Compliance (78.4/100) with critical gaps in automation bias safeguards and disaggregated performance metrics.
## Quick Start
### Prerequisites
- Python 3.12+
- Node.js 18+
- An Anthropic API key
### 1. Clone and Setup
```bash
git clone https://github.com/Mougatsu/claudehackaton.git
cd claudehackaton
```
### 2. Create `.env` file
```bash
cp .env.example .env
# Edit .env and set your ANTHROPIC_API_KEY