Platform Thinking: A Lemonade Case Study // Orr Shilon // MLOps Coffee Sessions #79

Lemonade's ML Engineering Team Lead, Orr Shilon, discusses the company's machine learning platform, emphasizing automation and efficiency. Lemonade leverages machine learning to improve user experience and business outcomes, such as predicting lifetime customer value. A key component is a Slack bot named Cooper, used for training and deploying models, reflecting the company's priority on automation driven by metrics like customers per employee. Lemonade built its own feature store before Tekton was available, customizing it to their specific data needs. The team consists of four people supporting over 20 data scientists, a ratio achieved through platform thinking and decentralized development.

Outlines

Part 1: Introduction, Context

Part 2: Automation, Monitoring

Part 3: Technical Stack, Infrastructure

Part 4: Team Culture, Talent

Part 5: Operations, Testing

Part 6: Strategy, Conclusion

Sign in to continue reading, translating and more.

Continue

MLOps.community

Part 1: Introduction, Context

Introduction to Orr Shilon and Lemonade's Advanced ML Platform

Lemonade's Custom Feature Store and the Build vs. Buy Conundrum

Orr Shilon's Bio and Lemonade's Use of ML in Insurance

Machine Learning's Role in Improving Lemonade's Product and Business

Part 2: Automation, Monitoring

Custom Monitoring Configuration and Feature Recycling at Lemonade

Automation Priority and the Slackbot's Role at Lemonade

Challenges in Automation and Decentralized Development at Lemonade

Part 3: Technical Stack, Infrastructure

Lemonade's ML Development Stack: Feature Store, MLflow, and Infrastructure as Code

The Biggest Challenge: Enabling Data Scientists with Varying Experience Levels

Balancing Openness and Structure in Lemonade's Model Serving Framework

Part 4: Team Culture, Talent

Platform Thinking, Talent Attraction, and Ownership in Data Science

Hiring Challenges and the Impact of VC Funding on ML Engineering Talent

Team Size, Support Ratio, and Platform Thinking at Lemonade

Infrastructure Leverage and Efficiency in Machine Learning

Part 5: Operations, Testing

Building Blocks Revisited: Airflow, Aporia, and the Model Serving Framework

Testing Challenges and the Need for Manual Data Scientist Involvement

Development Velocity, Accuracy, and Visibility: The ML Platform Mission

Part 6: Strategy, Conclusion

Early Engineering Involvement and Platform Thinking at Lemonade

Business Point in Time: Aligning Training Data with Prediction Times

Passion for Data Standardization and Closing Remarks

Platform Thinking: A Lemonade Case Study // Orr Shilon // MLOps Coffee Sessions #79

MLOps.community

Part 1: Introduction, Context

00:00Introduction to Orr Shilon and Lemonade's Advanced ML Platform

Introduction to Orr Shilon and Lemonade's Advanced ML Platform

02:31Lemonade's Custom Feature Store and the Build vs. Buy Conundrum

Lemonade's Custom Feature Store and the Build vs. Buy Conundrum

05:21Orr Shilon's Bio and Lemonade's Use of ML in Insurance

Orr Shilon's Bio and Lemonade's Use of ML in Insurance

07:40Machine Learning's Role in Improving Lemonade's Product and Business

Machine Learning's Role in Improving Lemonade's Product and Business

Part 2: Automation, Monitoring

12:04Custom Monitoring Configuration and Feature Recycling at Lemonade

Custom Monitoring Configuration and Feature Recycling at Lemonade

14:52Automation Priority and the Slackbot's Role at Lemonade

Automation Priority and the Slackbot's Role at Lemonade

18:11Challenges in Automation and Decentralized Development at Lemonade

Challenges in Automation and Decentralized Development at Lemonade

Part 3: Technical Stack, Infrastructure

20:27Lemonade's ML Development Stack: Feature Store, MLflow, and Infrastructure as Code

Lemonade's ML Development Stack: Feature Store, MLflow, and Infrastructure as Code

23:38The Biggest Challenge: Enabling Data Scientists with Varying Experience Levels

The Biggest Challenge: Enabling Data Scientists with Varying Experience Levels

25:37Balancing Openness and Structure in Lemonade's Model Serving Framework

Balancing Openness and Structure in Lemonade's Model Serving Framework

Part 4: Team Culture, Talent

27:45Platform Thinking, Talent Attraction, and Ownership in Data Science

Platform Thinking, Talent Attraction, and Ownership in Data Science

30:36Hiring Challenges and the Impact of VC Funding on ML Engineering Talent

Hiring Challenges and the Impact of VC Funding on ML Engineering Talent

33:45Team Size, Support Ratio, and Platform Thinking at Lemonade

Team Size, Support Ratio, and Platform Thinking at Lemonade

35:24Infrastructure Leverage and Efficiency in Machine Learning

Infrastructure Leverage and Efficiency in Machine Learning

Part 5: Operations, Testing

37:21Building Blocks Revisited: Airflow, Aporia, and the Model Serving Framework

Building Blocks Revisited: Airflow, Aporia, and the Model Serving Framework

39:32Testing Challenges and the Need for Manual Data Scientist Involvement

Testing Challenges and the Need for Manual Data Scientist Involvement

41:40Development Velocity, Accuracy, and Visibility: The ML Platform Mission

Development Velocity, Accuracy, and Visibility: The ML Platform Mission

Part 6: Strategy, Conclusion

44:46Early Engineering Involvement and Platform Thinking at Lemonade

Early Engineering Involvement and Platform Thinking at Lemonade

47:11Business Point in Time: Aligning Training Data with Prediction Times

Business Point in Time: Aligning Training Data with Prediction Times

49:28Passion for Data Standardization and Closing Remarks

Passion for Data Standardization and Closing Remarks