Marvis Virtual Network Assistant Overview

Juniper Networks has upended IT operations with artificial intelligence, machine learning, and data science. And this is how we got to Marvis. But we didn't get there overnight.

Several years ago, we began sending millions of data points from network devices to train our AI in the cloud, which became the basis for our AI primitives. With our data science toolbox, the AI learned as we trained new models and created Marvis, a conversational assistant unlocking access to AI for your IT team. Marvis learned to proactively find trouble in your network, laying the foundation for the self-driving action framework for the wireless, wired, and WAN.

Marvis is always working, alerting you to potential issues before users notice and can identify the causes of bad user experiences. With Marvis, we've changed the way you operate networks, moving the paradigm away from CLI, log files, and deciphering dashboards to simply having a conversation with the network. For IT teams, troubleshooting is now as easy as asking, Marvis, any unhappy users today? Marvis is a helpful member of your team who is always learning and scales with your needs.

And as large language models rekindle excitement about AI, integrating chat GPT is a natural evolution of Marvis's skill set, providing quick answers to documentation. Now, Marvis not only fixes your network, but helps you access troves of knowledge about networking. Sign up for a demo today and see for yourself how Marvis can help.

So, let's actually look at how much of this is real. So this is ChatGPT connected to the Marvis conversational interface, and essentially, you know, being able to ask ChatGPT, hey, you know, can we actually leverage large language models in Marvis, right, in the conversational assistant. So here's a question.

We've actually piped this question through the open ChatGPT model out there, and so we're actually getting, you know, natural language generation. Marvis always had natural language understanding, and we obviously had all this data to arrive at the right answer, but for the first time, we're actually introducing now natural language generation. Previously, if you asked us a question, "how do you configure dynamic port profiles?", we'd throw up a bunch of links like this, like Google used to do. Now we're leveraging ChatGPT natively in Marvis, and so this is coming to you, to your dashboard fairly soon in the Q3 timeframe.

I'm super excited about the Zoom integration. This one is incredibly powerful in what we can do with Zoom. Basically, you know, native integration from Zoom, and Teams will follow the second half of this year, but Zoom now, if you are a Zoom customer, we can natively bring in Zoom data into your Mist cloud. You can actually compare and contrast it with the loss, latency, jitter you see out there that Zoom sees. Labeled data, as Bob said, labeled data is gold when it comes to AI, and this is labeled data.

For the first time, we have a cloud architecture in this industry that can handle the Zoom data, every user, every minute, every Zoom call, all the time. We're consuming this from organizations right now, and actually be able to even predict. Bob didn't go into the details of explainability and predictability.

The predictability is us building a model that even if you don't have Zoom calls at a site, can I tell you if, you know, the wireless clients or the wireless RSSI, maybe the WAN latency, the round-trip time, what is causing this kind of implications on Zoom? One of the best parts of the Zoom integration is it's natively integrated into the Marvis conversational assistant. So the Marvis conversational assistant, you could ask, "tell me what Zoom calls, you know, happened today?" at the scale of your organization. You know, I have one customer, they're doing about 30,000 Zoom calls a day.

Just one customer, small customer, 30,000 Zoom calls a day. You're able to just ask the question, who's doing Zoom calls and how is that experience? The green dot says those were great Zoom calls. The user said that was good, no problems.

But the best query, my absolute favorite query on this whole feature is basically saying, you know what, we have all these clients, but how do I know, how do I know if there was one bad Zoom call today, right? Just ask that question to Marvis. And Marvis says, aha, there were sites that had bad Zoom calls, Kumar's MacBook Pro had a bad Zoom call today. And you click on it and voila, for the first time in the networking industry, we're taking data from truly real-time data, applying AI on it, we're applying that Shapley model that Bob talked about, and being able to come up with an answer saying, hey, that Zoom call was bad because that user was roaming at the exact same time.

So we'd say, okay, let's triple check this and let's ask Marvis, hey, Marvis, was that really a user having a bad experience? Of course that user was having a bad experience. And it's our favorite, it's RF engineers' worst nightmare, is this interband roaming all the time that happens. It's now native in the dashboard.

So this is available today. If you're a Zoom customer and a Mist customer, connect with your account teams and we can make this happen. We're taking not just Zoom's perspective of it, but also the user's feedback.

When you have that Zoom call, hey, how was that Zoom call? Nobody ever answers that question, except when the call sucks, everybody's like, oh, yeah, that call was bad, right? We need that labeled data, right? To say all is well in all these other instances, but this one call wasn't good. If you get that labeled data, it's awesome. And then, again, this is another really, really good advancement from Marvis' perspective in terms of sort of Marvis' new actions.

I won't do a full deep dive on the new actions here. Tomorrow, the entire boot camp, there's a switching section there, there's a wireless session, brand new Marvis actions are coming to the dashboard. We want you to please participate in tomorrow's boot camp. It's going to be phenomenal. It's customers presenting with Juniper product teams, and you will see this is one of my new favorite Marvis actions. Basically, think of it this way.

Imagine all the IoT sensors, video cameras, all of that in our network today. What happens suddenly if one of those video cameras stops sending traffic? Who's watching it? If a thermostat stops sending data to its cloud, who's watching it? Marvis is watching it, and this is Marvis catching it, right? And so, this is game-changing. And then, last but not least, probably my favorite of all of these is this Marvis as a member of your team.

Marvis is coming to a Teams channel near you. And so, Marvis, actually, we've submitted to Microsoft for the app integration here, and as soon as it gets approved, this will be released. But you can invite Marvis to a conversation with your fellow team member and say, hey, you know, Ryan, what's up with this particular infusion pump in this hospital? Ask Marvis, right? So, Marvis can participate in a conversation within your own team.

And this is super cool. And, you know, whatever you could do on the VNA, for the most part, you know, you can actually do natively in this. So, and all of us, how many of you are Teams customers here? Everybody, right? Literally everybody. Who is not?

And so, this, now, don't have to log into the Mist dashboard. Marvis is sitting on your Teams channels with you, and it could stay there. It literally can stay on that Teams conversation and then just keep answering questions as you go through this.

You can troubleshoot switching issues, troubleshoot, you know, wireless issues, whatever summary you see today on the conversational assistant in the dashboard is coming to this Teams channel. So, super stoked about this. And, you know, the little button that I love is adding it to your team.

Marvis as a member of your team. So, this is what Bob and team have been working on. So, thank you very much for all of that, Bob, and the innovations on AI.

Last year, last year, oh, by the way, we have to give away some stuff. So, who is celebrating a birthday today or closest to today? Who's got it? Stand up, please. Just tell me a name.

Literally, are you celebrating a birthday today? You got to stand up. We have five people pointing at you. Yesterday. Alright. Happy birthday, and on behalf of Juniper, I have a Wi-Fi access point and a T-shirt and whatever else you want. Happy birthday to you.

So, I think legally, I can't say whatever else you want, but, you know, we'll erase that from the record. Last time, last year when we had this event, the best, highest voted session was the customer panel because this is truly where you hear the stories, right? The what, the why, the how customers have transformed their networks. And this one, we actually have mics set up there as well.

I'm going to have a few questions, but really, it's yours. If you're new to Juniper, if you're new to Mist, if you are an existing customer, but like something that one of the customers has said, walk up to a mic, you know, interrupt us, this is a conversation for all of us to have. That's why this is an intimate session.

So, without further ado, please bring up my customer panel. Come on up. ♪♪♪