We have been working on operationalizing ML for past few years at CapitalOne Bank and would like to share our experiences and lessons we learned in building an ML platform, in our talk we plan to cover:
- Self-Service for Data Scientists
- Treat models, policies & features as content, not software, and allow live updates to content
- Provide software engineering best practices to ML content(s)
- How to meet enterprise need at scale
- Lightweight services
- Re-use models, data, and business logic wherever possible
- Containerize software to simplify scaling
- Multi-layer abstractions
- Respond to real time events
- Keep data in close proximity
- Focus on low-latency communication and fast computations
- Architect high-reliability services

Some of the questions this session intend to answer:
- Every FinTech enterprise needs to operationalize ML but most of them don't know where to start, how to deliver and more importantly what not to do?
- What architecture choices to explore and what tools to build to satisfy demanding needs of a thriving data science organization.
- How can you build ways to include data scientists in the agile development process, leveraging their expertise in feature engineering while enabling them to take part in DevOps practices without needing full DevOps experience.