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

git clone https://github.com/Mougatsu/claudehackaton.git
cd claudehackaton

2. Create .env file

cp .env.example .env
# Edit .env and set your ANTHROPIC_API_KEY

3. Install Python dependencies

pip install -e .

4. Start the Backend (FastAPI)

uvicorn regulationcoder.api.app:app --host 0.0.0.0 --port 8000 --reload

The API will be available at http://localhost:8000 (Swagger docs at http://localhost:8000/docs).

5. Start the Dashboard (Next.js)

cd dashboard
npm install
npm run dev

The dashboard will be available at http://localhost:3000.

6. Run the Demo

1. Open http://localhost:3000/evaluate
2. Click "Load Demo" to populate the TalentScreen AI profile
3. Click "Run Evaluation" for instant deterministic compliance check
4. Click "Deep AI Analysis (Claude Opus 4.6)" for AI-powered deep analysis (~60-90s)

CLI Usage

# Run compliance check from command line
regulationcoder check --profile tests/fixtures/talentscreen_profile.json --regulation eu-ai-act-v1

# Verify audit trail integrity
regulationcoder verify-audit --log-dir ./audit_logs

Architecture

Regulation PDF → Ingestion → Parsing → Extraction (Claude Sonnet 4.5)
    → Judge Gate A (Claude Opus 4.6) → Formalization (Claude Sonnet 4.5)
    → Judge Gate B (Claude Opus 4.6) → Evaluation Engine
    → Deep AI Analysis (Claude Opus 4.6) → Compliance Report

Tech Stack

Layer	Technology
AI Engine	Anthropic Claude Opus 4.6 + Sonnet 4.5
Backend	FastAPI, Pydantic v2, Python 3.12
Frontend	Next.js 14, Tailwind CSS, shadcn/ui, Recharts
CLI	Click + Rich
Audit	SHA-256 hash-chained append-only log

Key API Endpoints

Method	Path	Description
POST	/api/evaluate/	Deterministic compliance evaluation
POST	/api/evaluate/ai-analysis	Deep AI analysis (Claude Opus 4.6)
GET	/api/regulations/	List loaded regulations
GET	/api/requirements/	List extracted requirements
GET	/api/rules/	List compliance rules
GET	/api/audit/logs	View audit trail
GET	/api/audit/verify	Verify hash chain integrity

Project Structure

src/regulationcoder/
  core/         engine.py, judge.py, ai_analyzer.py, pipeline.py, config.py
  models/       clause.py, requirement.py, rule.py, profile.py, evaluation.py, ai_analysis.py
  ingestion/    pdf_processor.py, html_scraper.py
  parser/       segmenter.py, citation_extractor.py, tree_builder.py
  extraction/   requirement_extractor.py (Claude Sonnet 4.5)
  formalization/ rule_generator.py (Claude Sonnet 4.5)
  rules/        eu_ai_act_v1/ (Articles 9-15, 53 rules)
  audit/        logger.py, chain.py (hash-chained audit trail)
  exporters/    json_exporter.py, html_exporter.py
  api/          FastAPI app + routers
  cli/          Click CLI

dashboard/      Next.js 14 frontend
  src/app/      Pages: evaluate, requirements, rules, audit, upload
  src/components/ ScoreCard, ComplianceChart, RuleResultRow
  src/lib/      api.ts (typed API client)

License

MIT