MLaaS - Machine Learning as a Service

At Capital One we have built MLaaS (Machine Learning as a Service) which is SaaS platform part of our Machine Learning Ecosystem, meant to democratize and accelerate ML across the company.


We had 3 main challenges which drove this innovation:


1) we have lots of ML use cases and most of our Data Science Teams are working in silos, as part of business verticals. While this decentralized mode of operation gives them autonomy, it also makes the discovery and re-use of algorithms across the company quite a challenge. We are aiming at offering an easy way to discover and re-use in-house built algorithms, at scale, leveraging AWS scaling cloud infrastructure, for all DS teams.


2) We are a financial institution and hence deal with a high level of scrutiny, compliance and governance - we are aiming at streamlining the model governance and ensuring a well managed operating platform to progress models to production.


3) We needed a nimble and scalable infra to move from building local model solutions to evolving these to shared enterprise model microservices accelerating ML via reuse.


The result is MLaaS - a shared library of 50+ internally proven ML solutions across the company all built on AWS managed services. Key use cases are Fraud prevention, Cyber Anomaly detection, Model/Feature/Data pipeline monitoring, Credit Risk, Auto Dealer Sales Shifts, multiple NLP applications with many more use cases in flight and coming.