GeNet: A Multimodal LLM-Based Co-Pilot for Network Topology and Configuration

Managing communication networks in enterprise environments is complex, time-consuming, and errorprone. Research on automating network engineering has mainly focused on configuration synthesis, while changes to the physical network topology are often overlooked. In this paper, we introduce GeNet, which combines visual and textual information to interpret and modify network topologies and device configurations based on user intents. We evaluated GeNet using scenarios adapted from Cisco certification exercises relevant to enterprise networks and a series of different topology image variants. Our results show that GeNet can interpret images of network topologies, including low-quality handwritten ones, which can reduce the workload of network engineers and accelerate the network design process. Moreover, GeNet demonstrates the ability to handle intents successfully, even with incomplete or varying quality input specifications.