quantumopt uses a Graph Attention Network trained on 10,240 quantum circuits to predict optimization potential, then passes to Qiskit's transpiler for hardware-specific compilation targeting IBM Brisbane.

Results on 41 real QASMbench circuits: - 34% average gate reduction - 28% average depth reduction - 0 circuits made worse - 82% GNN prediction accuracy

Happy to answer questions about the GNN architecture, training pipeline, or quantum compilation approach.