Requirements:

Production experience shipping deep learning models at scale - systems serving real traffic under latency constraints
Ability to make architecture-level decisions independently - model selection, training infrastructure, serving strategy - and explain the reasoning and tradeoffs
Experience designing ML systems with hard latency and throughput requirements, including optimisation decisions (quantization, pre-computed embeddings, batching strategies)
Strong fundamentals in deep learning: gradient dynamics, attention mechanisms, graph message-passing, sequence modelling
Track record of influencing technical strategy across teams - you don't just build, you shape direction
Python, PyTorch (or equivalent), distributed training, ML pipeline orchestration

Nice to Have:

Experience in FinCrime, fraud detection, AML, or regulated financial services
Experience with graph-based methods (GNNs, entity resolution, link analysis) in production
Foundation model fine-tuning or LLM evaluation experience
Experience establishing modern ML practices in organisations scaling their ML capabilities

Designing and shipping ML and deep learning models for financial crime detection - sequence-based, graph-based, attention-based - serving real-time decisions at Wise's scale
Defining the architecture strategy for how Wise applies modern ML to risk - which model families, which serving patterns, which training paradigms
Building the reusable end-to-end pipeline pattern - from experimentation through training to production deployment - that future models follow
Evaluating and prototyping foundation model and embedding approaches for transaction representation across FinCrime domains
Partnering with Data Science on model evaluation, experimentation design and causal measurement in domains where clean A/B testing isn't always possible
Mentoring engineers and data scientists on modern ML fundamentals, production best practices, and architectural decision-making