Data Scientist, Ravelin

• Our ideal candidate is pragmatic, approachable and filled with knowledge tempered by past failures
• Experience building and deploying ML models using the Python data stack (numpy, pandas, sklearn)
• Understand software engineering best practices (version control, unit tests, code reviews, CI/CD) and how they apply to machine learning engineering
• Strong analytical skills
• Being a strong collaborator with colleagues outside of your immediate team, for example with client support teams or engineering
• Being skilled at communicating complex technical ideas to a range of audiences
• The ability to prioritise and to manage your workload
• Being comfortable working with a hybrid team

Desirable

• We are currently looking for a Data Scientist for a 1 year FTC to help train, deploy, debug and evaluate our fraud detection models
• Evaluating fraud models is hard; often times we do not even get labels for 3 months
• You’ll need to use your judgement when investigating cases of ambiguous fraud and when you’re investigating the veracity of the model itself
• We have to build robust models that are capable of updating their beliefs when they encounter new methods of fraud: our clients expect us to be one step ahead of fraud, not behind
• You will be given the equipment, space and guidance you need to build world class fraud detection models
• The work is not all green field research
• The everyday work is about making safe incremental progress towards better models for our clients
• The ideal candidate is willing to get involved in both aspects of the job – and understand why both are important
• You will be joining the Detection team
• The Detection team is responsible for keeping fraud rates low – and clients happy – by continuously training and deploying machine learning models
• We aim to make model deployments as easy and error free as code deployments
• Google’s is our bible
• Our models are trained to spot multiple types of fraud, using a variety of data sources and techniques in real time
• The prediction pipelines are under strict SLAs, every prediction must be returned in under 300ms
• When models are not performing as expected, it’s down to the Detection team to investigate why
• The Detection team is core to Ravelin’s success
• They work closely with the Data Engineering Team who build infrastructure and the Intelligence & Investigations Team who liaise with clients
• Build out our model evaluation and training infrastructure
• Develop and deploy new models to detect fraud whilst maintaining SLAs
• Write new features in our production infrastructure
• Research new techniques to disrupt fraudulent behaviour
• Investigate model performance issues (using your experience of debugging models)