Twitch link: https://www.twitch.tv/videos/740791498

Very practical, I’m feeling the same.

Author: Francesco Gadaleta

I do stuff with math, machine learning and programming.

• Founder & Software Engineer @ amethix.com
• Host @datascienceathome.com
• https://www.linkedin.com/in/fragadaleta
• https://medium.com/@frag

Data

• Upstream data update frequency
  • Validation data set from upstream team
• Data latency
  • 1 week old data, max allowed
• Real-time data ingestion?
• globalization - Regions?

Service performance

• Required TPS
• Request Latency
• Storage retention time
• DVC tool, data version control
  • https://dvc.org/

Science

• Model accuracy %
• Arruracy improve vs profit improve
  • ROI
• Model management?
  • Cuberflow
  • MLflow
  • AirFlow

Business

• Success measure for project
• Important business metrics from up/down stream teams
  • Used for talk with stakeholders
  • Debate with data

Team

• Simple structure is gold
• Coding standard
  • Especially scientist and engineer and economist
• Configuration standard template, across within team Remove fun part. Fun means risk.
• Data scientist (with no engineering workflow concept) will need 2 engineers support
  • so 1 applied scientist is enough
• Put people in the same room