Observe uses generative AI to simplify log explanations

[This transcript was auto-generated]
Keith Shaw 0:00
Hi, everybody, welcome to DEMO the show in which we have companies come in and they show off all of their new features and new technologies right here on the show. Joining me today is Tom Batchelor, he is the field CTO of Observe. And you guys have added some new generative AI features to your observability platform. Correct? That's right. All right. But before we get started, tell me about Observe and kind of what it provides for IT groups.
Tom Batchelor 0:21
Absolutely. So we're observe probably no surprise, we build an offer observability platform. So when we're building the platform, you know, it's no secret to anybody. All these applications we have running out there, they're generating a ton of data, logs, metrics, traces, other data that sits around that, whether that's build systems, source control systems, etc. So we saw the observability problem as a data problem. Okay, so we built a data platform to bring all of this data together. So we can then start to slice and dice correlate this data and dig in and ready to go and investigate and figure out what's happening when when these applications had issues. Now, before you've added the generative AI features, which we're gonna get to in a moment, what would what would companies do if they saw something in the platform, and they needed to understand more about what was going on? Yeah, so there's a couple of ways that they would have done this. So a lot of use of this data is in dashboards, like we have up here, that only goes so far. Then if you have, say, an unknown unknown, you need to dig into the data, then you already kind of getting into maybe writing queries, maybe trying to do things like regular expressions to go and extract things from raw text strings. So it kind of took experts to really go and do those deep investigations, who really knew not just the tool, but the application as well.
Keith Shaw 1:39
Right. So now what what new generative AI features have you added, which we're going to show on the show?
Tom Batchelor 1:44
So the the focus for our GenAI, Yeah, is really on helping people kind of get answers to the questions that they want quickly. So there's some things in there like just general guidance on where you might want to go into tooling. And we'll look at some other things around before you would have been manually querying data. Now we can automate some of that to make that life easier. Okay, and who is this benefit? What like, what are you looking to help the users that are already using the platform? And you're just making it easier for them to understand what's going on? Or are you trying to bring new users and new people into the platform to help explain, it's really both? Yeah. So yes, there's features in that to go and help new users. And that's where we'll start with a demo. Well, kind of how do I get started and going, one of the things we found is we built this out, is this has actually been accelerated for the expert users as well. So even our internal users that observe, we're using these features, because it's just a quicker way to get right, right, right. And I think when we talked before the show, approximately how much time have users been saving through through the use of years. So I've got to give a good example that we'll see this show the demo. Things like automated regex extraction, you might spend 30, 40 minutes trying to go and construct that to pull data out of your string. Yeah, that now happens in seconds. And automatically, it's a lot less painful.
Keith Shaw 3:01
All right, cool. Let's go jump into the demo. Show me what features.
Tom Batchelor 3:06
So we'll just start here. So this is kind of what we have here is just a dashboard showing my application health and things that are going on. I see this graph here, there's kind of looking at some errors, etc. Now, you know, if I could do this dashboard framework, new this application, they would know how to dive in, but let's pretend that we don't. So I'm going to pull up Olly GPT here. And I if I'm a brand new user, I might say, How can I view my container logs? I was just gonna go ahead and ask the Olly bot that question. So that comes here. So step one, login while we've done that, okay, let's go click on logs, as it says here. And we'll move this over. I searched for datasets container logs, we already have container logs expected, selected, so we're in a good spot. Okay. So now I'm looking at all my logs and our kind of standard features of the platform, I might try and do some kind of filters in here. So let's say you know, but I've got errors, I want to go and have a look at the access logs on my web server.
Now, I'll double click on one of these. So this is what we have. We have this string, there's URLs, response codes in IPs, etc. There's a couple of things I could go do. For one, I could say, Hey, can you explain this message. And this is really around that use case where I might take an error, I might then go into say, copy, paste, put it into Google, then go to Stack Overflow, etc. Well, let's short circuit that. Okay, and try and get the answers. So this is a basic example where we see here, okay, this is my response code. This is kind of the the URL I'm heading IPs, etc. Now, that's just one logline. That's not that useful. Let's say we want to break this down into columns. Maybe we want to kind of visualize this, do some filtering, etc. Sure. So if I click on here, I could say extract from string. Now, back to that regular expression example. I could go and figure out and type out that regular expression, but I'm going to hit Generate here. Now this takes a few seconds to run right but don't forget we will spend a lot of time.
Keith Shaw 5:00
This would take about 30 minutes if you were doing this on your own?
Tom Batchelor 5:02
It depends on the skill. Yeah, how good you are, how good you are. Yeah, this is a big one where we use this all the time, just because it's less painful. And here we go. This is the this is a regular expression. So something pretty complex here. Okay, right, this just got generated. We see here with examples of status code, where even assigning the correct type to this fields, which can be useful depending on what filters we want to do. So as I apply this, now we have all of these new columns. So I'm going to break this out, look at the past four hours, so that we can do some slightly different things. So I can visualize, and I couldn't do this with a raw text. But I can visualize this and break this out by status code.
And pretty quickly there, we get my visualizations, etc, I might decide, okay, so I'm fundamentally healthy, most things are being successful. So we'll just put a quick filter in here. We'll just for the first time Arason, where we have 500, or more spot status code. Okay. So now I'm looking here, I see. Okay, there's some sporadic issues and those things. So I might go back and have a look at the data. Now, I kind of have a question like I'm having an impact, but which areas of the site are impacted? Okay, which URLs Am I impacting? Right? Now, this is where previous world you would get into doing query language directly. What I can do here is I can go to my OPL console. And so this is the query we have here that we've been building through the UI. Now we kind of want to step beyond what the UI can do and kind of get into it. Yeah, but I don't need to know Opal. So if I want to know which URLs are impacted, I might want to do count by URLs as a comment here.
And then when I hit, I'm going to click a button, and we'll see here, this copilot, he's going to light up. And now we're using JPA, to go and write my query for me. Wow. Okay, so here we have that query. And now I kind of have these numbers against all these URLs. But we're kind of have arbitrary sorting here, right, probably want to know what's most impacted. So let's do sort, descending. And we'll do the same thing again.
Here we go. Now, I kind of have my query. So we can go execute that. Okay, and now I can see which particular pages which particular API's, etc, are most impacted by issues we have going on.
Now, of course, we would then go dive deeper into this and go into a whole bunch of other features. That's clearly obviously not the focus of today. Right. So, again, is it is it more about just getting the users to understand what's going on? Or is it more about the efficiency of getting them to do all a lot of these things quicker? So today, it's about the efficiency of helping people go and do the investigations they're doing today and do that quicker. Okay. And with less pain.
Keith Shaw 7:43
And were you utilizing either an open AI large language model, did you develop your own based on all of the data that you had internally? Or?
Tom Batchelor 7:52
So great question. So there's actually a mix in here. Okay. So things like the when we're engaging with the Olly GPT, and the little kind of chat window, yeah, things like where we're doing the regular expression extraction. So that's leveraging GPT. Things like the copilot we have here with a language that's our own model that we have trained on, essentially, the the code that we add in the opal code that we have within Observe.
Keith Shaw 8:14
And is this the first step that you're using with with generative AI? Or do you have future plans to integrate even more this is the this is the first step? Yeah. A big project that we have right now is something else we've noticed when companies go through these investigations, is for different sets of circumstances, there tends to be the same sets of steps that they go through and the same sets of kind of questions they asked if the data. So over time, we can learn those. So when we see similar kinds of failure signals, what are we actually going to be able to do is, rather than you looking at your 10, dashboards or your 10 queries or whatever, let's go run all those for you. And let's figure out which ones we think you should look at first. Okay, so that's generally going to help kind of remove the noise from from that investigation. Right. So what's the pricing model for these new features? Is it an add on? Or is it just built into the price of the observed platform? It's just built into the price website platform, there's nothing extra to get these features? And is it a subscription based model? Or do you use a different pricing method? So the model is its ingest based pricing?
Okay. All right. And where can people go for more information if they want to learn more about Observe or the Olly GPT?
Tom Batchelor 9:21
Observe Inc. is a great place to go additional content on that demos and blogs about how we're building this thing.
All right, Tom Batchelor from Observe AI. Thanks for joining us on the show. And thanks for the demo.