Senior AI/ML Engineer. Payments & Identity

You will lead applied research and production ML at the core of Aye’s identity and payments intelligence stack. This spans real-time biometric inference, fraud signal modelling, consumer behaviour learning, and multi-modal feature fusion, all under strict latency and privacy constraints. A defining part of the role is owning the A/B experimentation infrastructure that continuously validates model improvements in production, and contributing to the development of Aye’s small language model (SLM) for in-store personalisation and merchant intelligence.

What You’ll Build

• Train, optimise, and serve face recognition, liveness detection (active and passive), anti-spoofing, and presentation attack detection (PAD) models at sub-300ms p99 latency in production
• Design multi-modal inference pipelines that fuse biometric embeddings, transactional sequences, behavioural signals, and contextual metadata to power fraud detection and personalisation
• Build and operate A/B and multi-armed bandit experimentation frameworks in production; instrument feature flags, holdout groups, and statistical significance pipelines to drive continuous model improvement
• Develop and iterate on consumer behaviour models: purchase propensity scoring, cross-vertical affinity mapping, churn prediction, and next-best-offer recommendation for real-time CRM
• Contribute to Aye’s SLM initiative: fine-tune compact transformer models on anonymised retail interaction data for in-context merchant intelligence and consumer preference inference
• Optimise models for edge deployment using TensorRT, ONNX Runtime, and quantisation techniques (INT8, FP16) on NVIDIA Jetson and OpenVINO-compatible hardware
• Apply privacy-preserving ML techniques: federated learning, differential privacy, secure aggregation: to meet data residency and regulatory requirements across APAC jurisdictions

• 5+ years applied ML/CV in production systems, preferably in identity, fraud, or real-time payments scoring
• Strong Python engineering; production-grade experience with PyTorch or TensorFlow and their serving ecosystems (TorchServe, TF Serving, Triton Inference Server)
• Deep understanding of face recognition architectures (ArcFace, CosFace, AdaFace) and PAD/liveness detection techniques
• Experience designing and analysing A/B experiments at scale; familiarity with statistical methods (power analysis, CUPED, sequential testing)
• MLOps fluency: model versioning (MLflow, Weights & Biases), feature stores, data pipelines, model monitoring and drift detection in production
• Comfort with multi-modal learning: combining vision, NLP, and tabular signals in unified architectures
• Bonus: experience fine-tuning or distilling small language models (SLMs); knowledge of LoRA, QLoRA, or RLHF pipelines