Category Archives: machine intelligence

Towards ML Engineering: A Brief History Of TensorFlow Extended (TFX)

Much like Software Engineering followed Computer Science, as our world increasingly depends on machine learning systems, we argue that the time is right for ML Engineering to come of age. Our paper on ML Engineering is now available on arXiv. It covers a decade of production ML experience, illustrated by two successive end-to-end (E2E) ML platforms at Alphabet: Sibyl and TensorFlow Extended (TFX). I hope you will find it useful and that sharing our experience will help you advance your own ML systems.


Humbled to have played a small part in advancing both of these platforms working together with Konstantinos (Gus) Katsiapis, Abhijit Karmarkar, Ahmet Altay, Aleksandr Zaks, Alkis Polyzotis, Anusha Ramesh, Ben Mathes, Gautam Vasudevan, Irene Giannoumis, Jiri Simsa, Justin Hong, Mitch Trott, Noé Lutz, Pavel Dournov, Robert Crowe, Sarah Sirajuddin, Tris Warkentin, Zhitao Li and many other current and former Googlers acknowledged in this article.

Back to ML


After 5+ years working on Developer Relations @ Google I decided to go back to product development and joined the Research & Machine Intelligence team at Google as the Technical Program Manager. I’m working on machine learning (ML) infrastructure and large scale ML deployments, and could not be more excited about this next logical step of my journey.

Fun fact: my Computer Science undergrad degree was in AI, and I did neural network work during my graduate studies, but it took me a few years to get back into it. Those dang AI winters got in the way. Timing is everything.

More updates soon, in the meantime check out!

An obviously biased and highly subjective guide to Google Search at I/O 2015

The 100 days of Google Dev are in full swing so I thought I’d highlight a few of my favorite videos you may have missed. The list is obviously biased and highly subjective, but you already knew that. Over the past several quarters our Search team at Google has been busy adding capabilities that make it easier for mobile app developers to integrate with Google Search. Starting with App Indexing, through Voice Actions, the Knowledge Graph and Google Now on Tap, there are a lot of new products and APIs your app can integrate with. Why? The benefits include better user acquisition (read: app installs), more user engagement (read: rather than sitting idle, your app is actually used) and, last but not least, user delight (read till the end to see what I mean).

Let me walk you through what’s new.

Voice Interaction API

The Voice Interaction API is new in Android M. It allows your app to carry on a conversation with the user. While previously one could trigger an app using voice actions, this API takes it to the next level by allowing your app to maintain a dialog with the user. Learn more form this video by Sunil Vemuri and Barnaby James.

Search Console

Remember Webmaster Tools? No? That’s fine. Search released a new and improved version called Search Console, which lets you debug and analyze your Google Search performance. This includes troubleshooting App Indexing integration issues, which is very handy. To see how this all works, check out the video by Mariya Moeva.

AdWords, App Indexing and Analytics

While App Indexing brings organic user acquisition and engagement to your app, some people want to pay Google money, which I think is totally cool. Check out how you can do that through AdWords and how it all fits together by watching this overview by Jonathan Alferness and Scott Huffman. Both of them are VPs. You’ve been warned.

The Knowledge Graph

I have to admit that the Knowledge Graph is one of the coolest projects I got to work on at Google. To learn more about it check out my post on Medium, but here’s a new feature we’re rolling out – video actions and reviews. My colleague Satyajeet Salgar gives a nice overview in the video below.

Putting it all together

As you have probably figured out by now, Google Search has a number of products and APIs available for developers. In this video, Vivan Lum puts it all together. Vivan gets bonus points for the demo, so pay attention, yo.

Now on Tap

My favorite new Search feature announced at I/O is Now on Tap. It offers contextual assistance and leverages the Assist API, which is new in Android M. To see it in action check out the I/O keynote demo by Aparna Chennapragada. Developers can integrate with Now on Tap by implementing App Indexing.

If this isn’t enough videos for you to watch, check out the entire Google Search for Developers YouTube playlist and visit