Ilona:
Hello and welcome back to the Aviz Networks podcast series! I am your host, Ilona Gabinsky, and today we’re diving deep into our revolutionary GenAI-based Network Copilot with none other than Chid Perumal, the CTO of Aviz Networks. Chid, welcome to the show!
Chid:
Thank you, Ilona. It’s great to be here.
Ilona:
So Chid, AI now is the biggest change as it is fundamentally changing the relationship between the technology and humans, for the first time you can interact with the technology as another coworker. Can you tell us what makes Aviz’s Network Copilot so special in this evolving landscape?
Chid:
Great question, Ilona. I like the way you said fundamentally changing the relationship between technology and humans. In simple words, Aviz Network Copilot is an AI-powered assistant for your network. It is smart and proactive which evolves with your network and improves productivity by providing solutions for various use cases using a simple conversational interface. It’s like having a GenAI-skilled network engineer working alongside you, 24/7.
Ilona:
Oh, wow, that sounds impressive! And you know, I’m sure our listeners would love to know how they can actually start with it. How can they start integrating it into the existing infrastructure? Can you share some details?
Chid:
This is a great question. This is the first question which we ourselves asked, how can we introduce GenAI and create a product to help customers with that? What you need is a system here which is truly open. When I say open, it has to be ready to adopt to newer innovations coming and advancements coming in the LLM space, starting with picking the model for you because there are so many models available in the district. The second thing, it has to be interface study because LLMs are all about data. So the system should have a modern scalable architecture which is secure and robust. The most important thing, it should be able to easily integrate into your existing infrastructure because introducing GenAI should not affect your production network. That’s very important. And finally, since it is all about data, it should be able to work in a heterogeneous environment which involves various networking devices, switches, routers, various operating systems, and various tools which are used for managing the network because GenAI is all about data.
Ilona:
So as I understand, in order to start with a copilot, you need a system that works with any LLM that is enterprise-ready, vendor-agnostic, and can integrate easily with your existing infrastructure.
Chid:
Yes, that’s correct.
Ilona:
You know, and I’m sure it can solve a lot of problems for our customers. But can you just mention a few for us here?
Chid:
Yes, absolutely. Using GenAI, what we are trying to do with ours is to solve problems or the challenges for the use cases which require repetitive and mandated mundane tasks in a multi-vendor environment. For example, a few use cases I can list down is one is checking doing a network upgrade compliance. You need to do it because you are upgrading your network, whether it is a software upgrade or a hardware upgrade or a maintenance check. And after the upgrade, you need to make sure that the network’s operational behavior has not changed and this has to be repetitive and this has to be done over a period of time. After the upgrade, this can be simplified using GenAI. The second thing is security audits. It’s very important because security compliance is very important. Whether it can be config related or operational related, the IT team needs to continuously check your network to make sure that there are no conflict changes that have caused security violations in your network. The third one, measuring the network performance of your ISPs. We are primarily focused on performance monitoring of the ISPs because that is a true multi-vendor environment like you have routers and switches from different vendors and you have to correlate data against multiple data sources and also multiple tools. So these are some of the tasks which we believe GenAI can really help and solve.
Ilona:
This is really impressive to see how GenAI can eliminate repetitive and mundane tasks and let people focus more on innovation and strategy and things that make them really more efficient. But people also ask me all the time about prompt engineering. How does this whole prompt engineering work with Network Copilot?
Chid:
Yeah. Prompt engineering is very critical for Network Copilot. So if you ask me, prompt engineering is a well-defined process. First of all, it starts with a clear understanding of the problem which you are trying to solve because that’s the most important before you get into a solution, you need to know what you’re trying to sort. The second thing is you need to identify the data set and where you can collect this data set from. And then once you have identified this, then you have a huge amount of data, you have collected the data, you have a huge amount of data available. So you need to make sure that the solution provides accurate responses. So for that, technically speaking, you need to create the right SQL queries number one. And also you need to avoid inaccuracies. That is the hallucinations because models have to be fine-tuned to reduce the hallucinations. At Aviz, we use various techniques for helping with this. For the fine-tuning, we use something called a few-shot example, react prompting for accurate generation of the SQL queries. And for reducing hallucinations, we use something called the RAG approach by providing the context to the model so that the responses are aligned to your needs.
Ilona:
So basically, it’s like asking ChatGPT a question and it would not know unless it has the data and it has been trained and fine-tuned with that data.
Chid:
That is correct. It’s all about the quality of data needed for the fine-tuning.
Ilona:
Yes. So now for customers who already have a lot of network tools, how can Network Copilot easily integrate with them?
Chid:
Absolutely, Network Copilot is powered by the latest agent framework which is getting very popular in the GenAI space. With this approach, existing tools which are used for automation, network operations, and AIOps can easily be integrated. At Aviz, we have integrated our own NetOps tool ONES with Network Copilot for automatically creating templates for orchestration over a chat interface.
Ilona:
Wow. So this is really awesome what you’re saying that we can make all the existing tools powered by LLMs.
Chid:
Yeah, I believe that is where the future is heading towards.
Ilona:
Yes, definitely. And you know, I just came back from Cisco Live where I met with a lot of customers and by the way, they were really impressed with the demo with our Network Copilot demo that we showcased and they’ve been asking me some insightful questions and I would like to bring them up here so that we can answer them for our listeners here as well. And one of them, you know, is data security. So data security is a big concern. So how do we at Aviz ensure that the data is safe?
Chid:
Great question again. As I mentioned earlier, data security is one of the key tenets we have followed while developing Network Copilot. Network Copilot comes as a standalone package ready to be deployed in OnPrem environments without the need for internet access. What we say is we will never call an outbound API with the customers’ data so your data resides in your OnPrem itself and you control it. Secondly, as a product, Network Copilot also has enterprise-grade security features like role-based authentication, secure APIs for accessing and the accessing the data. So data store and also for the other interfaces.
Ilona:
Yeah. So I just want to emphasize again that this is really important for our customers to make sure that the data is safe for them and that it’s great that we can provide such capability. Another question that a customer has been asking: how do we actually connect to the data source?
Chid:
Connecting to data sources is straightforward with Network Copilot. The important requirement here is that data collection should not disrupt the existing infrastructure, that’s very important, right? Network Copilot provides a bunch of collection of data services which can easily integrate with standard mechanisms like SNMP, APIs, gNMI, and Flow Records. Also, we have built the injection pipeline in such a way that it can easily accommodate any custom tools we have with the customers.
Ilona:
Yeah, OK. Thank you for the insights. And here’s a specific use case question for you, Chid. A lot of customers are interested to see, for example, if their firewall is overloaded or not, is that something we help them with?
Chid:
Yes. Network Copilot can answer that question. But first, as I talked earlier about prompt engineering, we need to identify the data set and the context. So in this case, the Network Copilot can provide the response by looking at our network traffic data which is received by the firewall system, resource utilization, including this memory and hardware tables, and providing the context definition of what this overload means.
Ilona:
So basically, we can solve this use case for them, whether their firewall is overloaded or not. And what you are saying is that we can use GenAI to solve any problem based on customers’ requirements. So they can just build their own use case.
Chid:
Yeah.
Ilona:
And Chid, here’s one more question that people were asking, how does Network Copilot learn? Does it learn continuously?
Chid:
Yes, it does. Network Copilot uses continuous learning to adapt and evolve based on data it receives. Basically, we are talking about real-time data. Also, we are improving LLM responses by enforcing some guardrails to follow the interface-specific rules which can be legal, ethical, and make sure that the accuracy is always maintained, and this way we are learning and making Network Copilot better day by day.
Ilona:
Yes, this is great. And Chid, you know, I have the last question for you. This is a very interesting question. A lot of people were asking, will AI replace the engineer? And you being an engineer yourself, you know, what are your thoughts on this?
Chid:
Well, AI is a tool designed to augment human capabilities. That is what I believe in and not replace them. Of course, Network Copilot here helps you handle repetitive, routine, and time-consuming boring tasks, right? Allowing engineers to focus more on strategic innovations for their networks. It enhances productivity and empowers engineers to tackle complex challenges more effectively.
Ilona:
Yeah, I couldn’t agree more and I see it in my professional life also. AI definitely makes me more efficient but it cannot replace the creativity and innovation that humans bring. Thank you, Chid, for these insights. And before we wrap up, do you have any final thoughts for our listeners you would like to share?
Chid:
Yeah. Network Copilot by Aviz is a significant leap forward in how we manage and optimize networks. By leveraging GenAI, we are empowering businesses to achieve greater efficiency, reliability, and security for their network operations.
Ilona:
Thank you, Chid, thank you for your time, thank you for sharing your expertise. And thank you to our listeners for tuning in. Stay tuned for our next episode, where we’ll explore more about AI Fabric, data-driven network applications, and much more.
Chid:
Thank you.