As a robust analytical method, spectrofluorometric analysis with machine learning modelling has recently been used to authenticate wine from different regions, vintages and varieties. This preliminary study investigated whether the molecular fingerprint obtained with this approach is maintained throughout the winemaking process, along with assessing different percentages of wine in a blend. Monovarietal wine samples were collected at different stages of the winemaking process and analysed with the absorbance-transmission and fluorescence excitation-emission matrix (A-TEEM) technique. Wines were clustered tightly according to origin for the different winemaking stages, with some clear separation of different regions and varieties based on principal component analysis. In addition, wines were classified with 100 % accuracy according to varietal origin using extreme gradient boosting (XGB) discriminant analysis. The sensitivity of the A-TEEM technique was such that it allowed for accurate modelling of wine blends containing as little as 1 % of Cabernet-Sauvignon or Grenache in Shiraz wine when employing XGB regression, which performed better than partial least squares regression. The overall results indicated the potential for applying A-TEEM and machine learning modelling to wine chemical traceability through production to guarantee the provenance of wine or identify the composition of a blend.