Robot swarms rely on fast and reliable dissemination of control messages from a central orchestrator. Yet, achieving this over low-power wireless links is challenging. Mobility, interference, and link variability make some robots much harder to reach than others. We introduce Plyo, a relay-based flooding strategy in which only a small, carefully chosen subset of robots rebroadcasts orchestration messages. Selecting such relays is a difficult combinatorial problem, so we propose PlyoNet, a lightweight AI-based model that computes relay selections in a single forward pass. Across several swarm configurations, Plyo consistently matches a greedy relay selector in flooding performance while being orders of magnitude faster, achieving 2 ms inference on a MAX78000 microcontroller versus 440 ms for greedy on a microcontroller CPU. These results show that real-time, near-optimal relay-based TSCH flooding is feasible even on resource-constrained swarm orchestrators.