Efficient intelligence,
from silicon to quantum

A multi‑layer routing engine that eliminates GPU inference for routine queries. Combines persistent knowledge slots, fuzzy similarity cache (SimHash + Hamming distance ≤3), and a self‑learning PRL layer. Achieves verified 100% GPU cost reduction on a 200‑query test set even after deleting 70% of exact cache between runs.

First demonstration of passive rule extraction on a real quantum processor (ibm_fez). Learns process behaviour without disturbing the computation – foundation for efficient quantum characterisation.