⚖️ Legal AI Applications

# Legal Contract Analyzer - Neuro-Symbolic

from transformers import pipeline
import networkx as nx

class ContractAnalyzer:
    def __init__(self):
        self.nlp = pipeline("zero-shot-classification")
        self.legal_kg = self._build_legal_kg()

    def _build_legal_kg(self):
        """Build KG of legal concepts and risks"""
        g = nx.DiGraph()

        # Favorable terms for buyer
        favorable = ['limitation_of_liability', 'cap_damages', 'warranty_disclaimer']
        # Unfavorable terms for buyer
        unfavorable = ['indemnification', 'liquidated_damages', 'unlimited_liability']

        for term in favorable:
            g.add_node(term, favorability='high')
        for term in unfavorable:
            g.add_node(term, favorability='low')

        return g

    def analyze_contract(self, contract_text, buyer_perspective=True):
        """
        Analyze contract for risks
        """
        # Step 1: Extract key terms using NLP
        key_clauses = self.nlp(
            contract_text,
            candidate_labels=[
                'limitation of liability',
                'indemnification',
                'warranty',
                'payment terms',
                'termination',
                'confidentiality'
            ],
            hypothesis_template="This document contains: {}",
            multi_class=True
        )

        # Step 2: Score each clause
        risk_score = 0
        findings = []

        for label, score in zip(key_clauses['labels'], key_clauses['scores']):
            if score > 0.7:  # Confident detection
                node_data = self.legal_kg.nodes.get(label, {})
                favorability = node_data.get('favorability', 'neutral')

                if favorability == 'low' and buyer_perspective:
                    risk_score += 1
                    findings.append({
                        'clause': label,
                        'risk': 'HIGH',
                        'reason': f'{label} is unfavorable to buyer'
                    })
                elif favorability == 'high':
                    findings.append({
                        'clause': label,
                        'risk': 'GOOD',
                        'reason': f'{label} protects buyer'
                    })

        # Step 3: Generate report
        report = {
            'overall_risk': 'HIGH' if risk_score > 2 else 'MEDIUM' if risk_score > 0 else 'LOW',
            'findings': findings,
            'recommendations': self._generate_recommendations(findings)
        }

        return report

    def _generate_recommendations(self, findings):
        """Generate lawyer recommendations"""
        recommendations = []

        high_risk = [f for f in findings if f.get('risk') == 'HIGH']
        if high_risk:
            recommendations.append(
                f"Address {len(high_risk)} high-risk clauses before signing"
            )

        return recommendations