DOI: https://doi.org/10.36347/sjet.2026.v14i01.002

Pages: 10-40

AI-enabled chemistry is rapidly moving from single-instrument modeling to multimodal chemical measurement, where spectroscopy, chromatography–mass spectrometry, electrochemical sensing, and chemical imaging jointly constrain the same underlying chemical state. This review synthesizes how data fusion (low-, mid-, and high-level) can unify heterogeneous signals from molecules to materials and devices, and why physics guidance is essential for models that remain reliable under matrix effects, instrument drift, and cross-laboratory transfer. We discuss practical fusion architectures, hybrid mechanistic–learning models, and uncertainty-aware inference that converts predictions into decision-ready measurement results. A central theme is that multimodal AI must be evaluated as an analytical procedure: calibration, figures of merit, cross-modality consistency checks, and uncertainty budgets must be reported with the same discipline expected in analytical chemistry. We map common AI tasks by modality (peak deconvolution, spectral unmixing, retention-time prediction, MS annotation, EIS parameter estimation, image segmentation) and show representative case studies spanning pharma/biomedicine, food/environmental sensing, operando catalysis, energy devices, and polymer/materials quality control. Finally, we outline future directions: standardized multimodal benchmarks, interoperable metadata and formats, real-time closed-loop experimentation, greener miniaturized platforms, and trustworthy AI practices that support regulatory acceptance and deployment. Overall, multimodal measurement is converging toward a new paradigm: quantitative, uncertainty-aware, and deployable chemical inference from fused evidence rather than isolated instrument readouts.