Q-Nidaan

A quantum machine learning platform for clinical screening. Upload a biomedical dataset, it compresses the features down to a qubit budget, trains a hybrid quantum-classical model, and reports held-out performance next to a classical model trained on exactly the same features.

Four model families are trained and compared: a variational quantum classifier with its own gradient-descent loop, a quantum-kernel support vector machine, a projected quantum kernel, and a hybrid quantum neural network over medical images. Every run is written to versioned JSON with area under the ROC curve, specificity at 95% sensitivity, training time, conformal intervals and calibration flags, plus five-seed sample-efficiency curves.

The interesting part is the honesty. On the datasets benchmarked so far, including renal genomics with 54,675 genes compressed to eight qubits, a classical model matches or beats the quantum one on six out of six, and the platform says so in its own README. Inference was also run on real hardware: IBM ibm_marrakesh, a 156-qubit machine, with dynamical decoupling, measurement twirling and readout mitigation, and the predictions agreed with the simulator four out of four.