GenAI is the talk of the town. In all sectors, from IT to healthcare to retail shop floors, discussions are dominant about the way it can drive meaningful changes. One of the domains to see merit in the technology early is networking. The number of off-the-shelf, AI-based networking software already existing speaks to that.
It was expected that at this Networking Field Day event, the audience will be treated to several such products. One company, however, stood out, not just because of the word “GenAI”, but for ahead of the curve thinking. Aviz Networks recently launched a new solution called Network Copilot which they showcased at the event. It is the crowning piece of Aviz Networks’ Networking 3.0 stack.
“Network Copilot is a top-down approach,” says Chid Perumal, Co-Founder and CTO.
Capturing Information from Data
Lately there has been a significant ground swell in network data that enterprises are desperate to tap into. Unfortunately, the growing crop of cookie-cutter data analytics tools that allow users to harness portions of that data with some AI software, cannot deal with that influx, nor cater to the wide array of emerging business use cases.
To unlock the trapped value in data, enterprises need an AI solution that can blend in with the existing stack without gaps or spaces – and allow anyone to apply GenAI without the groundwork. Network Copilot fits that description squarely.
A Personal AI Assistant for the Network
Network Copilot is a smart AI assistant that can answer any questions about the network – think network compliance, configuration, business analytics, capacity planning, and state anomalies.
But it is more than just a conversational chatbot. It is a full-blown system that works by optimizing and tuning the network from within, and can be personalized to deliver bespoke outcomes.
Network Copilot is not another point solution that you can deploy out of the box and make the stack bigger, says Perumal. It is a system in itself, one that is robust, flexible and secure.
Based on pre-trained open-source LLMs that constantly learn from the network, Network Copilot furthers the makers’ goal to push open source advancements deeper into the enterprise.
Perumal provided a glimpse of the architecture – data is ingested from cloud or on-prem. “It’s not a single ingestion pipeline that you get choked. It’s a cluster so that it can scale horizontally.”
Data is either stored in Aviz Networks’ own database that comes as part of the service, or written to a customers’ database. It then goes through an orientation pipeline where it is cleaned, normalized to a single base, and correlated. RAG is implemented to build and refine the knowledge base. At the end of the pipeline is an API Gateway to expose the results. Alternatively, users can get information by asking the Aviz chat interface questions.
Enterprise-ready with all the integrations in place, Network Copilot boasts broad compatibility with switches, ASICs, network operating systems (NOS), and more. Use of pre-trained community LLMs allows it to unlock significant cost savings compared to proprietary solutions – no need to hire an internal team of data scientists.
Getting Started
Getting started entails a couple steps which begin with preparation. Pre-deployment, the teams need to define the set of use cases, and identify the datasets for the job. Aviz Networks advises customers to do the customary cost- benefit analysis at this stage, to examine the pros and cons.
“If you can do something within 5 minutes, and GenAI can get you down to 4, it is not useful,” says Perumal.
On deployment, the model needs to be trained on network data by integrating with pre-existing data lakes, or generating its own. Contexts must be imported to make effective correlations, and conclusions about causation.
Network Copilot has universal compatibility built into its fabric. It is out and out vendor-neutral, meaning it can ingest data coming from any source – any switch or NOS.
“Today, if you look at any network infrastructure, on-prem or cloud, there are solutions which collect data – monitoring frameworks, capabilities that are built-in already,” he says. “We just use the data source, not the actual devices so that it is less disruptive.”
Aviz Networks’ LLM is as versatile, as it is forgiving. Deployment is undemanding – users can choose their own preferred hardware ensuring choice, control and cost optimization. “You can do an inference on GPU as well as on CPU,” he says.
Targeted Enterprise Use Cases
Highlighting the enterprise use cases that Network Copilot targets, Perumal points out four specific ones. The first on the list is network compliance. Network Copilot offers persona-focused information on network compliance.
Another star capability is capacity planning and forecasting. Network Copilot provides visibility into the current utilization and helps predict future demands. Aviz Networks works with the customer network architecture team on strategic resource management and planning to ensure continuity of performance and scalability.
Other use cases include troubleshooting and on-demand business analytics. Perumal notes, “Today to do troubleshooting, NetOps people go back to various dashboards, and look at alerts. We have to simplify this with LLMs” because quick diagnosis is half the battle won.
Network Copilot allows users to fashion their own custom dashboards that highlight only the analytics of interest specific to the organization. This way, additional noisy information that sidetracks the important things can be eliminated. The dashboards will only provide curated information and logic that accelerate decision-making.
For maximum accuracy and efficiency, the model needs to go through prompt engineering where datasets are tuned, RAG is implemented, and context is fed to ensure little to no hallucination.
For more, be sure to check out Aviz Networks’ demo of Network Copilot from the recent Networking Field Day event.
