Hi HN. I built this bare-metal C++ engine to handle mass-node failures in orbital mesh networks. Like During a simulated 300-satellite failure, exact algorithms like Dijkstra take over 400ms to calculate a new route, which drops the connection. I used an ALNS metaheuristic to find a route with a 2.1% optimality gap in just 49ms. It then naturally converges to the 0% global optimum in the background in few seconds all this in standard CPU Core. I’d love to hear your thoughts on the destroy/repair operators I used.