What is GuifiSants mesh network?

Barcelona's community WiFi net — thousands nodes, volunteer-run, real urban grit for 20+ years.

How to train GNN on my own mesh data?

Grab the Kaggle notebook, swap your topology CSV for node features/edges, retrain on T4 GPU in hours.

Does this GNN beat standard routing protocols?

94% delivery prediction vs. baselines' 78%; excels on flaky links, per held-out tests.

This GNN Eats Barcelona's Messy Mesh Data — And Thrives

Barcelona's rooftops hide a gritty truth: mesh networks are wild beasts, not tidy simulations. One team's GNN, fed 31 days of real GuifiSants data, finally tames them.

Open Source Beat Apr 11, 2026 3 min read

Visualization of GuifiSants mesh network graph with GAT attention weights highlighting flaky links

⚡ Key Takeaways

Real GuifiSants data — 31 days, 7,931 samples — crushes synthetic graphs for GNN training. 𝕏
GAT with 4 layers, 8 heads learns link nuances like battery life and interference automatically. 𝕏
94.2% accuracy; ONNX export enables fast Rust inference on edge devices. 𝕏

Published by

Open Source Beat

Community-driven. Code-first.

#GAT Models #GAT Routing #GAT Training #GNN Training #Graph Neural Networks #Mesh Networks #Real Data AI #Real Data Training #Real-World AI Data #Real-World Data

Worth sharing?

Get the best Open Source stories of the week in your inbox — no noise, no spam.

Originally reported by Dev.to

⚡ Key Takeaways

The 60-Second TL;DR

Open Source Beat

Share this article

Worth sharing?

Related Stories

The 3AM Satellite Scare That Sparked Probabilistic Graph Neural Networks for Anomaly Hunting

Hjarni Gives OpenClaw Agents the Memory They've Been Missing

300+ Destructive Endpoints Lurking in Your AI Agent's Toolbox

Scrapy Maintainer Drops Bombs on AI Scrapers: Code's a Breeze, Pages Fight Back

Stay in the loop