Hungarian FX Mortgage Legal Research System

System architected, designed, programmed & ideated by https://www.bencium.io/

⚠️ IMPORTANT LEGAL DISCLAIMER

🚨 THIS SOFTWARE IS NOT FINANCIAL OR LEGAL ADVICE 🚨

By using this open-source software, you acknowledge and agree:

• This system provides research assistance only, not professional legal or financial advice
• You use this software entirely at your own risk and responsibility
• Bencium and contributors are NOT LIABLE for any damages, losses, or legal consequences
• All generated documents must be reviewed by qualified legal professionals before use
• This software does not guarantee legal outcomes or financial recovery
• You assume full responsibility for any legal actions based on this system's output

⚖️ Professional Review Required: Generated legal documents should always be reviewed by qualified legal professionals before use in actual legal proceedings.

---

📖 Background | Háttér

🇬🇧 English

The Problem:

Between 2006-2016, Hungarian banks sold CHF (Swiss Franc) and EUR-denominated mortgages to hundreds of thousands of consumers without properly explaining the exchange rate risk. When the Hungarian forint devalued, monthly payments doubled or tripled. Many borrowers lost their homes.

2025 - The Turning Point:

• April 2025: CJEU ruled in C-630/23 that entire FX contracts can be declared invalid if the bank didn't clearly explain FX risk
• July 2025: Hungarian Supreme Court (Kúria) issued Uniformity Decision 10/2025, aligning Hungarian law with EU precedent
• ~43,000+ victims joined consumer advocacy groups seeking justice

What This System Does:

This AI-powered legal research assistant helps lawyers rapidly process FX mortgage cases using the new 2025 precedents:

1. OCR documents - contracts, correspondence, payment history (via Mistral AI)
2. Extract violations - 9 clause types including FX risk, broker liability, insurance breaches
3. Match to precedents - CJEU rulings (C-630/23, C-186/16, C-260/18, C-705/21, C-520/21)
4. Calculate damages - exchange losses, fees, compound interest, opportunity cost
5. Generate legal documents - complaints, motions, settlement demands

Why Rust?

Reason	Explanation
Performance	Legal research needs <500ms responses for precedent matching
Reliability	Banking/legal data requires strict correctness - Rust catches errors at compile time
Single binary	Easy deployment to lawyers' machines without runtime complexity
Local-first	All data stays on user's machine (SQLite + Chroma) - critical for attorney-client privilege
Memory safety	No crashes when processing thousands of documents

---

🇭🇺 Magyar

A Probléma:

2006 és 2016 között a magyar bankok CHF (svájci frank) és EUR alapú jelzáloghiteleket értékesítettek több százezer fogyasztónak anélkül, hogy megfelelően tájékoztatták volna őket az árfolyamkockázatról. Amikor a forint leértékelődött, a havi törlesztőrészletek megduplázódtak vagy megháromszorozódtak. Sokan elvesztették otthonukat.

2025 - A Fordulópont:

• 2025. április: Az Európai Unió Bírósága (CJEU) a C-630/23 ügyben kimondta, hogy a teljes devizahitel-szerződés érvénytelenné nyilvánítható, ha a bank nem tájékoztatta egyértelműen az ügyfelet az árfolyamkockázatról
• 2025. július: A Kúria 10/2025. számú jogegységi határozatával a magyar jogot az EU precedensekhez igazította
• ~43.000+ károsult csatlakozott fogyasztóvédelmi szervezetekhez

Mit Csinál Ez a Rendszer?

Ez az MI-alapú jogi kutatási asszisztens segíti az ügyvédeket a devizahitel-ügyek gyors feldolgozásában a 2025-ös precedensek alapján:

1. Dokumentumok OCR feldolgozása - szerződések, levelezések, fizetési előzmények (Mistral AI-val)
2. Jogsértések kinyerése - 9 záradéktípus, beleértve az árfolyamkockázatot, közvetítői felelősséget, biztosítási jogsértéseket
3. Precedensek illesztése - CJEU ítéletek (C-630/23, C-186/16, C-260/18, C-705/21, C-520/21)
4. Károk kiszámítása - árfolyamveszteségek, díjak, kamatos kamat, elmaradt haszon
5. Jogi dokumentumok generálása - keresetek, beadványok, egyezségi ajánlatok

Miért Rust?

Ok	Magyarázat
Teljesítmény	A jogi kutatásnak <500ms válaszidő kell a precedens-illesztéshez
Megbízhatóság	A banki/jogi adatok szigorú pontosságot igényelnek - a Rust fordítási időben észleli a hibákat
Egyetlen bináris	Egyszerű telepítés ügyvédi gépekre futtatókörnyezet nélkül
Helyi adattárolás	Minden adat a felhasználó gépén marad (SQLite + Chroma) - kritikus az ügyvéd-ügyfél titoktartáshoz
Memóriabiztonság	Nincs összeomlás több ezer dokumentum feldolgozásakor

---

🤖 NEW: AI-Powered Case-Agnostic System

🏛️ Intelligent legal research assistant for ANY Hungarian foreign-currency mortgage case

This system has been completely transformed from a hardcoded demo into a production-ready, AI-powered legal research assistant that can intelligently analyze ANY Hungarian FX mortgage case - regardless of bank, currency, or loan type.

✨ NEW AI-Powered Features

🧠 Intelligent Case Analysis

• Universal Bank Support: Works with ANY Hungarian bank (Erste, Aegon, OTP, K&H, CIB, Raiffeisen, UniCredit, etc.)
• Multi-Currency Handling: Analyzes CHF, EUR, USD, JPY, GBP, HUF loans automatically
• Dynamic Document Processing: AI understands any contract format or structure
• Multi-Case Support: Handles bank switching and multiple concurrent loans

🤖 AI-Driven Capabilities

• Mistral AI Integration: Advanced document understanding and legal analysis
• Dynamic Damage Calculation: AI calculates damages based on YOUR specific case facts
• Personalized Legal Documents: AI generates case-specific complaints and submissions
• Real-Time File Monitoring: Automatic reanalysis when new documents are added
• User Override System: Correct any AI misinterpretations with easy interface

🎯 Core Features

• 🔍 Document Analysis: OCR processing of ANY mortgage documents via Mistral AI
• ⚖️ Legal Clause Extraction: AI identifies violations specific to YOUR contracts
• 📚 Precedent Matching: Intelligent matching against CJEU and Hungarian court cases
• 💰 Financial Calculator: Case-specific damage calculations (no hardcoded values!)
• 📄 Legal Document Generation: Personalized complaints, PBT submissions, lawyer letters
• 🌐 Bilingual Support: Complete Hungarian and English language processing
• 🔒 Privacy-First: OCR uses Mistral (cloud) for scanned pages; analysis receives only anonymized text

🏗️ System Architecture

Hungarian FX Mortgage Legal Research System
├── 🤖 AI Analysis Engine (NEW)
│   ├── src/ai/           # AI-powered analysis
│   │   ├── mistral_client.rs  # Mistral AI integration
│   │   ├── case_analyzer.rs   # Intelligent case analysis
│   │   └── file_watcher.rs    # Real-time file monitoring
│   └── src/api/case_analysis.rs # Dynamic API endpoints
├── 🦀 Rust Backend
│   ├── src/api/          # REST API endpoints
│   ├── src/extractors/   # Contract clause extraction
│   ├── src/matching/     # Legal precedent matching
│   └── src/models/       # Data structures
├── 🐍 Python OCR Processor
│   ├── mistral_client.py # Mistral AI integration
│   └── main.py          # Document processing pipeline
├── 📊 Local Databases
│   ├── SQLite           # Case metadata and analysis
│   └── Chroma           # Vector similarity search
└── 🌐 Dynamic Web Interface # AI-powered UI

📤 Browser Uploader + Privacy (EN / HU)

🇬🇧 English

What users should know (browser-only queue):

• Local queue: Files are stored in your browser (IndexedDB) so you can add from multiple folders over time.
• Explicit approval: OCR does not start until you check the approval box and click Start OCR.
• Cloud OCR only: Scanned pages are uploaded to Mistral OCR for text extraction. This is the only step where page images leave your device.
• Local PII scrub: PII detection + anonymization happens locally in the browser (human‑in‑the‑loop).
• Masked analysis only: Only anonymized text is sent to Mistral for legal analysis. Raw text is never sent.
• Confirmation required: The “Continue with Analysis” button stays disabled until you confirm you reviewed anonymization.
• Supported formats: PDF, TXT, EML, MSG, HTML/HTM.
• Clear local queue: Click New Files or clear site data in your browser to remove stored files.

🇭🇺 Magyar

Amit a felhasználónak tudnia kell (böngészős sor):

• Helyi várólista: A fájlok a böngészőben (IndexedDB) tárolódnak, így több mappából is hozzáadhatók később.
• Kifejezett jóváhagyás: Az OCR nem indul el, amíg be nem jelölöd a jóváhagyást és rá nem kattintasz az OCR indítása gombra.
• Felhős OCR: A szkennelt oldalak a Mistral OCR-hez kerülnek szövegkinyerésre. Ez az egyetlen lépés, amikor az oldalképek elhagyják az eszközt.
• Helyi PII‑szűrés: A személyes adatok felismerése és anonimizálása helyben, a böngészőben történik (ember a folyamatban).
• Csak anonimizált elemzés: Az AI‑elemzéshez csak anonimizált szöveg kerül elküldésre, nyers szöveg nem.
• Megerősítés kötelező: A „Continue with Analysis” gomb csak az anonimizálás ellenőrzése után aktív.
• Támogatott formátumok: PDF, TXT, EML, MSG, HTML/HTM.
• Helyi sor törlése: Kattints a New Files gombra, vagy töröld a webhelyadatokat a böngésződben.

💾 Database Schema

Local SQLite Database:

• legal_cases - CJEU and national court precedents
• documents - Processed contract documents
• extracted_clauses - Identified contract violations
• precedent_matches - Legal precedent similarity scores
• financial_calculations - Damage and restitution amounts
• generated_reports - Complete legal analysis packages

📋 Prerequisites

1. 🦀 Rust (1.70+) - Install from rustup.rs
2. 🐍 Python (3.8+) - For OCR document processing
3. 🟢 Node.js (18+) - Required for local server - Download from nodejs.org
4. 🔑 Mistral API Key - Get from platform.mistral.ai
5. 📄 Tesseract OCR (Optional) - For fallback document processing

🚀 Quick Start

🤖 AI-Powered System (NEW)

# 1. Clone repository
git clone https://github.com/bencium/devizahitel-copilot.git
cd devizahitel-copilot

# 2. Configure Mistral API (REQUIRED for AI features)
cp .env.example .env
# Edit .env and add your Mistral API key

# 3. Start the AI-powered system
./start_ai.sh

# 4. Open http://localhost:8080 
# ✅ AI-powered interface ready for ANY case!

📖 Complete Setup Guide → SETUP.md

For detailed instructions, troubleshooting, and advanced configuration, see SETUP.md

🎯 How It Works - AI Workflow

1. Add Your Documents

# Place your documents in the OCR output folder
cp your_contracts/*.pdf ocr_output/
# System automatically detects new files!

2. AI Analysis Process

1. Open: http://localhost:8080
2. Click: "Analyze Case with AI" button
3. AI Processing:
   • 🤖 Identifies ALL banks and currencies in your case
   • 📄 Extracts loan amounts, dates, payment history
   • 💰 Calculates personalized damage amounts
   • ⚖️ Matches relevant legal precedents
4. Review & Correct: Override any AI misinterpretations
5. Generate: Download personalized legal documents

3. User Override System

• Problem: AI extracted wrong amount?
• Solution: Use Corrections tab to fix it
• Result: System immediately recalculates everything!

4. API Integration

# Health check
curl http://localhost:8080/health

# Get legal precedents  
curl http://localhost:8080/api/cases

# Process document
curl -X POST http://localhost:8080/api/research/analyze \
  -F "file=@contract.pdf" \
  -F "language=hu"

Legal Precedents Database

The system comes pre-loaded with key CJEU precedents:

1. C-186/16 Andriciuc v Banca Românească (2017) - Information disclosure requirements
2. C-520/21 Szcześniak v Bank M. (2023) - No bank compensation for invalid contracts
3. C-705/21 MJ v AxFina Hungary (2023) - Full restitution for unfair terms
4. C-26/13 Kásler v OTP Bank (2014) - Transparency requirements
5. C-630/23 ZH, KN v AxFina Hungary (2025) - Latest precedent on contract invalidation

Clause Extraction Patterns

The system recognizes several types of clauses:

• FX Risk Clauses: Foreign currency risk allocation terms
• Transparency Clauses: Information disclosure requirements
• Interest Rate Clauses: Variable interest rate terms
• Penalty Clauses: Fees and penalty provisions
• Unfair Terms: General unfair contract terms

Generated Legal Documents

The system can generate:

• Complaints: Initial court filings citing relevant precedents
• Motions: Legal motions for contract invalidation
• Appeals: Appellate briefs with precedent analysis

Development

Adding New Precedents

Add cases to the precedent database by modifying:

// src/db/mod.rs - get_default_precedent_cases()

Adding New Language Support

1. Add language patterns to src/extractors/language_detector.rs
2. Add clause extraction patterns to src/extractors/clause_extractor.rs
3. Update the language detection logic

Extending Clause Types

1. Add new patterns to src/models/clause.rs
2. Update extraction logic in src/extractors/clause_extractor.rs
3. Add matching logic in src/matching/precedent_matcher.rs

Testing

# Run all tests
cargo test

# Run with logging
RUST_LOG=debug cargo test

# Test specific module
cargo test --package devizahitel_legal_research --lib models::case::tests

Performance Considerations

• Database Indexing: Full-text search indexes on case text and clause content
• Caching: Consider Redis for frequently accessed precedents
• Embeddings: OpenAI API calls are rate-limited; implement local caching
• OCR Processing: Tesseract can be slow; consider background job processing

🏗️ Architecture & Development

System Architecture, Design, Programming & Ideation by Bencium.io

This comprehensive legal research system represents innovative AI-first architecture combining:

• Advanced legal document processing with Mistral AI
• Sophisticated precedent matching algorithms
• Comprehensive financial damage calculation models
• Privacy-first local deployment strategy

📄 Legal Disclaimer

This system is designed for legal research assistance only. Generated legal documents should be reviewed by qualified legal professionals before use in actual legal proceedings.

Contributing

1. Fork the repository
2. Create a feature branch (git checkout -b feature/new-feature)
3. Commit your changes (git commit -m 'Add new feature')
4. Push to the branch (git push origin feature/new-feature)
5. Open a Pull Request

License

This project is licensed under the MIT License - see the LICENSE file for details.

Support

For issues and questions:

1. Check the existing GitHub issues
2. Create a new issue with detailed description
3. Include logs and reproduction steps

Roadmap

• Machine learning-based clause classification
• Integration with legal databases (Westlaw, LexisNexis)
• Advanced document comparison features
• Client portal with case management
• Mobile application
• Integration with court filing systems