Hi! 👋 I'm Mario.

I’m a Senior Machine Learning engineer at Intercom.

I help Intercom run production machine learning services.

Our systems solve problems for customers. They have to be fast and reliable. They hit all types of edge cases users type in.

That means that our solutions can’t not work.

⚡️ On this site, I write about all the things that made our machine learning team faster.

Everything I wish I had when we’ve formed the team. Tools that made us reinvent the wheel fewer times, processes we’ve found refreshing, and lessons I wish someone had written for me when I started building systems heavily based on real-world data.

🧠 Why am I the guy you should learn about this stuff from?

Before I joined the machine learning team, I had worked on the biggest scaling problems in Intercom. It equipped me with principles to build simple, resilient systems that need to change all the time (Intercom deploys new versions 100+ times a day).

Since 2017, I’ve been actively exploring how to build machine learning systems with these principles.

My experience is quite broad. I’ve built systems from latency sensitive predictions as you type to bulky ETLs that crunch terabytes of data to extract insights into conversational data.

Before I joined Intercom, I had worked in Microsoft and Palantir.

🍳 Some non-ML stuff about me

I’m originally from Croatia, but I currently live in Dublin, Ireland 🍀 And I enjoy it!

I spend lots of free time cooking, running, and cycling. Not great at following recipes, which makes my life spicier than prescribed. Still not sure if I’m a 🐈 or a 🐕 person.

If you want to connect, I tweet about work and life on @mariokostelac.