With CMOS technology scaling for miniaturization and performance improvement, analog and radio frequency integrated circuits (RFICs) are very much affected by process variations leading to performance degradation. A post-manufacturing healing mechanism with low overhead is desirable to compensate for these variations. A machine learning (ML)-assisted sensing mechanism has been proposed in this work for design of nonintrusive process variation sensor (PVS). Low-cost test signatures (TSs) extracted from these PVS are placed close to the circuit under test (CUT) and this does not load or couple to any nodes of CUT. As the TSs remain invariant with the shifts in control knobs (CKs), performance healing against process variations can be achieved with a single set of extracted TSs. A novel one-step healing approach is proposed in this work for selection of appropriate CK with the help of a neural regression model during the production test in automated test equipment (ATE). The proposed healing approach is illustrated on a current-starved voltage-controlled oscillator (CSVCO) circuit, to recover yield loss due to fabrication process variations.