Are we deployed yet? Turning AI research into a revenue engine

Description

Enterprises are making large investments in integrating AI/ML into their business, and yet it remains challenging to transform these initiatives into core revenue-driving functions. The key reason for this gap is that AI and ML have traditionally been research initiatives dominated by prototypes and proof of concepts, whereas operationalizing AI requires packaging, testing, setting up data pipelines, and enabling monitoring—areas squarely outside the scope of a researcher. As a result, research models often get shipped to a software engineering team tasked with reimplementing the model without proper context. It therefore takes many weeks to even months to deploy a model into production.

As creators of ModelDB, an open source model management solution developed at MIT and the proprietary Verta platform, Verta.ai has managed and deployed several hundreds of models ranging from cutting-edge deep learning models for autonomous driving to traditional ML models in finance and retail. Across these diverse application areas, Manasi Vartak has found common practices and tools that are crucial to productionizing AI/ML models. She examines key processes and tools by which enterprise AI teams can leverage and think about model data pipelines, A/B testing, reproducibility, and interpretation to drive revenue. She also demonstrates how these practices can equip data-driven enterprises to cross the chasm between research and production.