Software Engineer, Intuit
Sumanth is a Software Engineer at Intuit, where he focuses on building scalable ML services. He has previously built Big data and ML systems at Verizon and Apple. Sumanth holds a Masters in Computer Science from the University of North Carolina at Charlotte.
The presentation introduces Intuit AI Model Monitoring Service (MMS). MMS is an in-house Spark-based solution developed by Intuit AI to provide ongoing monitoring for both data (statistics of model input/output etc.) and model metrics (precision, recall, AUC etc.) of in-production ML models. The project is soon to be open-source. MMS aims to tackle multiple challenges of in-production ML model monitoring:
- Integration of multiple data sources from different time ranges: in order to generate all metrics to monitor an in-production model, we often need to integrate multiple datasets with different schema from different time range. For example, in order to compute model metrics like AUC, the collected ground truth is always collected in a different data set with a few days or even months delay after we record the model's output data. In other cases, we might need to integrate additional dimensional data so that we can create different segments to analyze the model per segment.
- Reusable and extendable metric and segmentation library: it is not scalable to develop a metric/segmentation logic per model. How to create a reusable yet extendable library to hold the metric and segmentation logic is a challenging task by considering different models might have distinct data schema. Model owners are able to take advantage of MMS to create and schedule pipelines without writing any code to monitor in-production models. MMS is able to integrate generic data and also provides a programming API to be fit into a specific data schema generated by a certain ML platform. MMS also allows developers to use MMS' APIs to create reusable metric and segmentation logic in an open-contribution library. MMS pipelines are very scalable and Intuit is using MMS to integrate 10M+ rows and 1K+ columns of in-production data to generate 10K+ metrics for in-production models.'