ORIGINAL ABSTRACT

Field-deployable food-safety monitoring requires sensing platforms that can detect chemical residues rapidly while also determining whether each measurement is reliable enough for decision-making. Here, we report an artificial intelligence (AI)-gated multiplex electrochemical sensing platform that integrates electrochemical impedance spectroscopy (EIS) with reliability-aware decision logic. A 16-electrode antibody-functionalized array was designed for simultaneous detection of eight residues: the nitrofuran antibiotic metabolites 3-amino-2-oxazolidinone (AOZ), 5-methylmorpholino-3-amino-2-oxazolidinone (AMOZ), semicarbazide (SC), and 1-aminohydantoin (ADH), together with the pesticides glyphosate, atrazine, paraquat, and chlorpyrifos. The platform was evaluated in shrimp, milk, chicken, salad, and soil-runoff extracts. Instead of treating each EIS spectrum only as a classifier input, impedance responses were converted into interpretable electrochemical features, including impedance magnitude, phase behavior, baseline state, spectral roughness, and dose-response distance. These features were assessed using AI gates for signal quality, baseline drift, dose-response consistency, and prediction uncertainty before final reporting. The sensor achieved strong analytical performance, with R2 ≈ 0.975, detection limits of 0.001-0.06 ng mL-1, recoveries within 80-120%, and readout within 5 min. Compared with conventional XGBoost classification, AI-gated reporting improved accepted-sample accuracy from 89.1% to 98.7% by routing uncertain or inconsistent spectra to retest/review rather than forcing binary safe/unsafe decisions.