Matt: Hello everyone. Welcome to LogRocket. My name is Matt Arbesfeld. I'm the CEO and co-founder here at LogRocket and I'm joined today by Renzo Lucioni. He has over 10 years professional experience working in the data analytics space. Graduated from Harvard with a CS degree and is the lead engineer working on Galileo, LogRocket's new machine learning feature set. Hey Renzo, welcome to the podcast. Renzo: Hey Matt. Hey guys. Excited to be here. Matt: Awesome. Maybe to start, you could share a bit of background and how you got into working on the Galileo project here. Renzo: Sure, I'd be happy to. So I've been an engineer at LogRocket for a little more than five years now. It's gone by really quickly and have had the opportunity to work on several different parts of the product in that time, including our session replay features and our issue reporting features. And I guess in so doing, saw an opportunity to combine those two streams of data to do something a little new that I've wished an error reporting tool could do for me for many years. And that is find the most important handful of problems in that big pile of problems or possible problems that you often see in traditional error tracking tools. Matt: Can you talk more about that? Of what problem did you face as an engineer using traditional error reporting tools in the past? Renzo: Generally what I've experienced using these error tracking tools is they are very good at collecting everything and you'll end up with a list of thousands or tens of thousands of possible problems that you can never reasonably hope to get through yourself. And if you could do that, there are some very compelling examples of your users suffering and your application buried in there, you're going to have a hard time finding them and they may go unreported or you may force your users to reach out to you repeatedly and you may never end up finding out about it. And it would be nice if something somehow could pull that small handful of really meaningful, impactful problems out from that big stack so that you can then use your time more effectively and go work on those and save time by not having to wade through everything and triage. Matt: Yeah, that makes sense. I remember setting up an error monitoring tool years ago and I set it up, we pushed production and I went to sleep and I woke up in the morning and the slack was filled with thousands and thousands of errors and you're like, "Where do I start with all of this?" So that seems amazing. I guess maybe at a high level, how do you discern what's an important error versus which one is not like that seems like it would be a challenging problem to figure out. Renzo: It definitely is a challenging problem. That's something that we've been thinking a lot about over the course of the last year or so. It turns out that really helpful indicator in determining whether some problem is impacting the user or not is looking at how the user's reacting when that problem occurs. It turns out that people generally behave in similar ways when they're surprised by something, when something doesn't work the way you were expecting it would. And vice versa, if a problem is silent, has had no impact on your experience and you're not surprised, you're going to react in a different set of ways. And what we've been able to do is start to separate these two groups from each other and use that information to find those issues that actually impact people's experiences and suppress the ones that don't. Matt: Yeah, I was booking a hotel room like a week ago and I was trying to choose the location picker and I remember I kept clicking again and again and again. And so are those the kind of user reactions you look at in terms of users clicking a lot or refreshing the page, I guess? Renzo: Those are good examples. There may be more subtle things like how you're moving your mouse or how you are scrolling up and down the page. And this strategy that we've taken is using machine learning to, instead of enumerate these possible patterns ourselves, present all this information to the machine learning algorithm and let it do the job of finding those patterns that are most effective at creating the separation that we're after. Matt: And finally, what do you see as the long term plan for these machine learning features? Where does this go in 2, 5, 10 years? Does AI just write all the code for me itself or where can this get to? Renzo: I don't know about writing all code, fixing all the plugs for you, at least not anytime soon, but ideally, you should be for example, alerted when one of these likely to be impactful problems has occurred. And generally just have confidence that this short list of problems is a good representation of what your engineering team should really be paying attention to fixing versus getting distracted by a lot of other stuff that could take up your time. I think general speaking, that's where I see it going. Matt: No, that makes sense. A single important issue that is affecting customers for days could be very costly and you may have to hire teams of people to watch over all this data. So to be able to do that automatically is super powerful. Renzo: Exactly. Matt: Renzo, thanks for sharing so much and for anyone who's interested to learn more, you can visit the link down in the description and sign up for the Galileo beta list and we'll also be sharing more of the upcoming months as the beta rolls out to more and more customers. So thanks everyone for the time today and hope to hear from you soon.