Vicki Boykis — Machine Learning Across Industries

Production Hurdles

Vicki Boykis highlights the intricate challenges of deploying machine learning models into production. She notes that the process is more complex than traditional software deployment due to the need for data management, model drift planning, and service orchestration 1. Vicki explains that prototyping and solidifying steps are crucial, as packaging models often involves creating REST endpoints or using Docker containers 2.

Putting stuff in production is really hard. And so I would say that's the biggest thing.

--- Vicki Boykis

These challenges underscore the importance of thorough planning and understanding of both data and model intricacies.

   

Metadata Management

Managing metadata is a significant yet often overlooked aspect of data science projects. Vicki emphasizes that many companies struggle with metadata management, which is crucial for updating models and conducting analyses 3. She mentions that while open-source tools like Amundsen are emerging, there is still no single solution for comprehensive metadata management.

People actually clamor for that, more so than even visibility into how to manage the model.

--- Vicki Boykis

Without standardized metadata, companies face issues like not knowing which data is proprietary or how to efficiently query data lakes.

   

Team Structures

The structure of data teams significantly impacts their effectiveness. Vicki discusses the benefits of both centralized and embedded data science teams, noting that smaller companies may benefit from a centralized approach, while larger companies might find embedded teams more effective 4. She highlights the risk of siloed teams leading to duplicated efforts, emphasizing the importance of collaboration.

I've seen it work well different ways in different companies.

--- Vicki Boykis

Ultimately, the choice between centralized and embedded teams depends on the company's size and specific needs.