Design Patterns MCP Server

{ "patterns": [ { "id": "bias-mitigation", "name": "Bias Mitigation Pattern", "category": "AI Governance", "description": "Systematically identifies, measures, and reduces bias in AI systems", "problem": "AI models trained on biased data produce discriminatory outcomes", "solution": "Implement bias detection and mitigation strategies", "when_to_use": ["Training AI models for decision-making systems"], "benefits": ["Fair outcomes", "Regulatory compliance"], "drawbacks": ["Increased complexity"], "use_cases": ["Hiring systems", "Loan approval"], "related_patterns": ["fairness-pattern"], "complexity": "High", "tags": ["ai-governance", "ethics", "bias"], "code_examples": [ { "language": "python", "code": "import pandas as pd\nimport numpy as np\n\nclass BiasDetector:\n def __init__(self, protected_attribute='gender'):\n self.protected_attr = protected_attribute\n\n def detect_bias(self, X, y, protected_attributes):\n privileged_mask = (protected_attributes[self.protected_attr] == 1)\n unprivileged_mask = (protected_attributes[self.protected_attr] == 0)\n \n privileged_rate = y[privileged_mask].mean()\n unprivileged_rate = y[unprivileged_mask].mean()\n \n disparate_impact = unprivileged_rate / privileged_rate if privileged_rate > 0 else 0\n \n return {\n 'disparate_impact': disparate_impact,\n 'privileged_rate': privileged_rate,\n 'unprivileged_rate': unprivileged_rate\n }\n\ndetector = BiasDetector()\nbias_metrics = detector.detect_bias(X, y, protected_data)\nprint(f'Disparate Impact: {bias_metrics[\"disparate_impact\"]:.3f}')" } ] }, { "id": "fairness-pattern", "name": "Fairness Pattern", "category": "AI Governance", "description": "Ensures AI systems provide equitable outcomes across demographic groups", "problem": "AI systems can perpetuate societal inequalities", "solution": "Integrate fairness metrics and constraints into ML pipelines", "when_to_use": ["Building decision-making AI systems"], "benefits": ["Equitable treatment", "Regulatory compliance"], "drawbacks": ["Accuracy-fairness trade-offs"], "use_cases": ["Automated hiring", "Credit scoring"], "related_patterns": ["bias-mitigation"], "complexity": "High", "tags": ["ai-governance", "ethics", "fairness"], "code_examples": [ { "language": "python", "code": "import numpy as np\n\ndef calculate_fairness_metrics(y_true, y_pred, sensitive_features):\n metrics = {}\n \n for group in sensitive_features.unique():\n mask = sensitive_features == group\n group_metrics = {\n 'selection_rate': y_pred[mask].mean(),\n 'accuracy': (y_true[mask] == y_pred[mask]).mean()\n }\n metrics[str(group)] = group_metrics\n \n if len(metrics) >= 2:\n selection_rates = [metrics[g]['selection_rate'] for g in metrics]\n metrics['selection_rate_disparity'] = max(selection_rates) - min(selection_rates)\n \n return metrics\n\nfairness_metrics = calculate_fairness_metrics(y_test, predictions, test_data['gender'])\nprint(f'Selection Rate Disparity: {fairness_metrics.get(\"selection_rate_disparity\", 0):.3f}')" } ] }, { "id": "interpretability-pattern", "name": "Interpretability Pattern", "category": "AI Governance", "description": "Makes AI model decisions transparent and understandable", "problem": "Complex AI models act as black boxes", "solution": "Implement interpretability techniques like SHAP values", "when_to_use": ["High-stakes decision-making systems"], "benefits": ["Increased transparency", "Easier debugging"], "drawbacks": ["Computational overhead"], "use_cases": ["Financial risk assessment", "Medical diagnosis"], "related_patterns": ["bias-mitigation"], "complexity": "Medium", "tags": ["ai-governance", "interpretability"], "code_examples": [ { "language": "python", "code": "import shap\nfrom sklearn.ensemble import RandomForestClassifier\n\nclass InterpretableModel:\n def __init__(self):\n self.model = RandomForestClassifier(random_state=42)\n self.explainer = None\n\n def fit(self, X_train, y_train):\n self.model.fit(X_train, y_train)\n self.explainer = shap.TreeExplainer(self.model)\n return self\n\n def explain_prediction(self, X_instance):\n shap_values = self.explainer.shap_values(X_instance)\n return dict(zip(X_instance.columns, shap_values[1][0]))\n\nmodel = InterpretableModel()\nmodel.fit(X_train, y_train)\nexplanation = model.explain_prediction(X_test.iloc[[0]])\nprint('Feature contributions:', explanation)" } ] }, { "id": "ai-privacy-pattern", "name": "AI Privacy Pattern", "category": "AI Governance", "description": "Implements privacy-preserving techniques in AI systems", "problem": "AI systems require access to sensitive personal data", "solution": "Apply federated learning and differential privacy", "when_to_use": ["Processing sensitive data", "Multi-party AI"], "benefits": ["Enhanced privacy", "Regulatory compliance"], "drawbacks": ["Increased complexity"], "use_cases": ["Healthcare models", "Financial risk assessment"], "related_patterns": ["bias-mitigation"], "complexity": "High", "tags": ["ai-governance", "privacy", "security"], "code_examples": [ { "language": "python", "code": "import numpy as np\nfrom sklearn.linear_model import LogisticRegression\n\nclass DifferentialPrivacy:\n def __init__(self, epsilon=1.0):\n self.epsilon = epsilon\n\n def add_noise(self, data, sensitivity=1.0):\n scale = sensitivity / self.epsilon\n noise = np.random.laplace(0, scale, data.shape)\n return data + noise\n\n def train_private_model(self, X_train, y_train):\n model = LogisticRegression(random_state=42)\n model.fit(X_train, y_train)\n return model\n\ndp = DifferentialPrivacy(epsilon=0.5)\nnoisy_data = dp.add_noise(X_train.values)\nprivate_model = dp.train_private_model(noisy_data, y_train)" } ] }, { "id": "ai-governance-pattern", "name": "AI Governance Pattern", "category": "AI Governance", "description": "Establishes comprehensive governance frameworks for AI systems", "problem": "AI systems lack standardized governance", "solution": "Implement governance framework with policies and monitoring", "when_to_use": ["Large-scale AI development", "High-risk applications"], "benefits": ["Regulatory compliance", "Risk mitigation"], "drawbacks": ["High implementation complexity"], "use_cases": ["Financial AI systems", "Healthcare applications"], "related_patterns": ["bias-mitigation"], "complexity": "High", "tags": ["ai-governance", "ethics", "compliance"], "code_examples": [ { "language": "python", "code": "import datetime\n\nclass AIGovernanceFramework:\n def __init__(self, organization_name: str):\n self.organization_name = organization_name\n self.policies = {}\n self.risk_assessments = []\n\n def define_policies(self, policies: dict):\n self.policies = policies\n\n def assess_risk(self, model_info: dict) -> dict:\n risk_score = 0.0\n if model_info.get('type') == 'deep_learning':\n risk_score += 0.5\n if model_info.get('critical_system'):\n risk_score += 0.3\n\n return {\n 'risk_score': risk_score,\n 'level': 'HIGH' if risk_score > 0.5 else 'LOW'\n }\n\nframework = AIGovernanceFramework('AI Corp')\nframework.define_policies({\n 'data_governance': 'Encrypt sensitive data',\n 'model_monitoring': 'Track performance metrics'\n})\n\nrisk = framework.assess_risk({'type': 'deep_learning', 'critical_system': True})\nprint(f'Risk Level: {risk[\"level\"]}')" } ] } ] }