Griffin, ML Platform at Instacart // Sahil Khanna // MLOps Podcast #145
MLOps.community
00:00
How to Implement a Real Time Pipeline?
Batch was good for us to test things very quickly. Then we moved to there was this bridge, which we may show like going to online service. Plus it was useful for us to handle those state queries, which we couldn't compute dirty patch. And then we introduced a production critical system. So the next upgrade was now we are using all the batch features in our future store, which are computing every day or every week. Now let's introduce more real time features, which are computed every minute or every few minutes. We call them on demand features which use something that comes from say like a search query or something that's only available at prediction time.
Transcript
Play full episode
