Deploying data science into production is a big challenge. With the rapidly changing available data sources, types, and the methods available for their analysis, there is a continuous need to update deployed data science models frequently. This makes it difficult to count on agreed –upon standards and design or work within the framework of exclusive tools. Envestnet | Yodlee delivers financial insights using transaction data to millions of users every day. Our ‘Transaction Data Enrichment’ solution turns ambiguous transaction information into clear, contextualized data using sophisticated artificial intelligence and machine learning to achieve industry-leading accuracy across a wide range of account and transaction types. This talk is not just about our journey towards successfully productionizing such a data science solution, but also about sharing our learnings and some best practices we have built over the years working with massive volume and ever changing schema of data. In this session, we will discuss the steps for building deployment ready data science solutions – from problem formulation, strategies for ground truth creation, etc. to model deployment, monitoring, maintenance and updates. We will also understand why measuring quality is a critical trade-off call in data science.
Chief Analytics Officer, Vice-President Data Sciences and Analytics, Envestnet Yodlee