The Network Data Analytics Function (NWDAF) is central to enabling zero-touch network management in fifth-generation (5G) networks by supporting real-time analytics and closed-loop automation. Despite its critical role, open-source NWDAF implementations remain limited in scope and accessibility. In this paper, we develop an open-source NWDAF, compatible with the open-source core network Free5GC, that collects network data via subscriptions to Network Functions (NFs), and also includes an integrated Large Language Model (LLM) interface that enables natural language interaction with human operators. The interface processes user intents, encodes them using a semantic embedding model, and maps them to one of seven predefined intent categories to trigger analytics queries or event subscription commands. This architecture abstracts the complexity of traditional interfaces, allowing non-expert users to manage network analytics and subscriptions with ease. The system supports Access and Management Function (AMF) and Session Management Function (SMF) event subscriptions, real-time monitoring, and analytics retrieval via Prometheus, all accessible through a conversational interface. By bridging AI-driven intent recognition with standardized network analytics, our implementation enhances operator usability and provides a foundation towards AI-native 6G networks. The source code and datasets generated during the current study are available in the github repository, https://github.com/HenokDanielbfg/testbed.
Alex: Welcome to another episode of ResearchPod.
Sam: Today we're looking at a paper that asks a genuinely interesting question about 5G networks—the kind of mobile network that powers your phone right now. These networks are extraordinarily complex to manage, and the researchers want to know: what if you could just talk to one, in plain English, instead of writing complicated technical commands?
Alex: So the idea is replacing specialist code with something more like a conversation?
Sam: Exactly. Right now, managing the core of a 5G network requires deep specialist knowledge of very specific technical interfaces. The researchers propose a system where an operator could ask a question in plain English and get useful data back, without writing a single line of code.
Alex: What makes the current setup so difficult in the first place?
Sam: 5G networks are built as a collection of independent services, each with rigid, specific digital pathways for communication. Think of it like a professional kitchen. You can't just walk in and order "dinner"—you have to specify the exact stove temperature, the precise timing of every ingredient, and which chef handles which station. The technical term for this structure is "Service-Based Architecture." It's powerful, but managing it manually demands expertise that most people simply don't have.
Alex: So the only way to get a meal right now is to already be a trained chef who knows every single step?
Sam: That's a good way to put it. To address this, the researchers built a system centred on a component called the Network Data Analytics Function—NWDAF for short. Think of it as a central control room that constantly watches the whole network. They then connected it to a Large Language Model—the same kind of technology that powers modern chatbots—to serve as the interpreter between what a human asks and what the network actually needs.
Alex: How does that interpreter actually work? Does it genuinely understand what I'm asking?
Sam: This is where the distinction matters. The system doesn't understand language the way a person does. Instead, it uses a process called semantic embedding. Imagine every possible sentence being plotted as a point on a vast map, where sentences with similar meanings end up close together. When you type a question, the system converts your words into a position on that map, then finds the closest pre-approved technical command. It's measuring how close your request is to something it already knows how to do—not genuinely comprehending it.
Alex: And that distinction matters for reliability, I'd imagine.
Sam: It does. They reinforce this with a technique called Retrieval-Augmented Generation, or RAG. Rather than letting the AI guess freely, the system first consults a pre-approved list of allowed actions—like checking a rulebook before acting. It retrieves the right procedure from that list, then generates the specific command. This keeps the system from inventing responses that the network can't actually execute.
Alex: So if I ask about a user's connection drop rate, the system finds the closest approved action, confirms it's valid, and pulls exactly the right data?
Sam: That's the idea. It maps your intent to the specific technical calls needed to retrieve that information. The goal is to make these large, complex systems accessible to people who aren't network specialists—which is a meaningful step forward, if it holds up at scale.
Alex: Which raises the obvious question. Does it hold up?
Sam: The authors are careful not to overclaim. The system performs well in their controlled test environment, but a real-world 5G network serving millions of devices simultaneously is a very different challenge. Traffic is unpredictable, devices are constantly moving, and edge cases multiply quickly.
Alex: How do they stress-test it, then?
Sam: They use what they call mobility models—essentially digital scripts that simulate a person's daily routine. A virtual commuter wakes up, travels across the city, stops at a shop, comes home. Run thousands of these simultaneously, and you get a realistic picture of how the network behaves under pressure—traffic spikes, sudden handovers between towers, unexpected load patterns—without needing a live network full of real users to experiment on.
Alex: So instead of waiting for a real person to accidentally walk into a coverage dead zone, they can simulate thousands of people doing exactly that, on demand.
Sam: Exactly. It gives researchers a repeatable, controllable way to find weaknesses before the system ever touches a real network.
Alex: That's a sensible sequencing. Build the conversational layer, then stress-test it against realistic human behaviour.
Sam: That's the arc of the research. The paper argues that the combination—a language model grounded by a pre-approved action list, sitting on top of a dedicated analytics hub, tested against realistic movement patterns—is a credible path toward networks that operators can manage through conversation rather than code. The concept is sound. The remaining work is proving it survives contact with the real world.
Alex: Thanks for walking us through that, Sam. And thanks to everyone listening to ResearchPod.