Senior Machine Learning Engineer

Payle is building financial infrastructure for the AI agent economy, enabling autonomous agents to transact under human-defined rules. The Senior Machine Learning Engineer will design and own risk, fraud, trust, and future credit-scoring systems for autonomous payment authorization, while establishing ML infrastructure, governance, explainability, and fairness standards.

Responsibilities:

• Design and deploy real-time transaction risk scoring for AI-agent-initiated payments, with a hard latency budget of 20ms at p99
• Build fraud detection models covering velocity anomalies, amount outliers, new-merchant risk, structuring patterns, and behavioral deviations from agent baselines
• Develop agent trust scoring: longitudinal models that evaluate an agent's historical transaction behavior, outcome verification rate, policy adherence, and dispute history
• Design and implement an offline/online feature store with point-in-time correctness guarantees, backed by our append-only transaction ledger
• Build feature pipelines from raw event streams (authorization, outcomes, chargebacks, policy violations, merchant interactions) with automated freshness and consistency validation
• Establish feature lineage, documentation standards, and automated leakage detection integrated into CI
• Train, evaluate, and deploy models starting with gradient-boosted trees (LightGBM/XGBoost) on tabular transaction data, evolving toward more complex architectures as labeled data volume supports
• Implement SHAP-based explainability for every production decision: per-transaction factor attribution with stable, auditable output
• Establish model versioning, registry, and staged deployment (shadow mode → canary → full production) with documented rollback procedures
• Build drift monitoring: population stability index (PSI) tracking per feature, automated alerting on distribution shift, and scheduled retraining pipelines triggered by drift thresholds
• Design behavioral underwriting models for agent-initiated credit (FundingFlex): probability of default, exposure at default, and loss given default models trained on agent transaction history
• Build adverse action reason code generation: every credit decision must produce machine-generated, human-readable justification codes compliant with regulatory requirements
• Implement disparate impact testing and fairness metrics (demographic parity, equalized odds) on all underwriting-adjacent models, with documented thresholds and review cadence

Requirements:

• 5+ years building and deploying ML systems in production, with at least 2 years on fraud detection, risk scoring, credit underwriting, or a related financial services domain
• Strong Python with production ML frameworks: scikit-learn, LightGBM/XGBoost, pandas, NumPy
• Deep understanding of point-in-time feature correctness, target leakage, and the failure modes of offline-online feature skew
• Experience with model explainability (SHAP, LIME, or equivalent) and the ability to translate model outputs into business-actionable insights
• SQL proficiency for feature extraction from transactional databases at scale
• Experience with model versioning, A/B testing infrastructure, and staged rollout strategies
• Master's or PhD in a quantitative field (Computer Science, Statistics, Applied Mathematics, Physics, Economics) or equivalent demonstrated experience
• Experience with on-device or edge ML deployment
• Familiarity with GDPR Article 22 (automated decision-making), ECOA/Reg B adverse action requirements, or EU AI Act risk classification
• Experience building ML systems for payment networks, digital wallets, or BNPL platforms
• Publications or open-source contributions in fraud detection, anomaly detection, or credit risk modeling