Lanthanum, as a fluorescent element, was encapsulated in complexing silica xerogel by using a one-pot procedure: silica formation from tetraethyl-orthosilicate and 3-aminopropyltriethoxysilane, its chemical modification by reacting with 2,4-dihydroxybenzaldehyde, and lanthanum complexation through the formed hydroxy-azomethine groups occurred in the same reaction system. A polydimethylsiloxane-a,x-diol with average number molecular weight of 20,000 was added in the sol-gel system to facilitate obtaining the flexible, free-standing films. Using different molar ratios among the reactants, a series of experiments was performed and the fluorescence of the resulted compounds was evaluated, since the photophysical properties of the prepared compounds strongly depend on the composition. The acquired data have been used for fluorescence modeling based on soft computing instruments. Neural network models, individually or aggregated in stacks, were developed to estimate the fluorescence intensity depending on the reaction mixture composition: tetraethyl-orthosilicate, 3-aminopropyltriethoxysilane, 2,4-dihydroxybenzaldehyde, lanthanum acetate, and polydimethylsiloxane-a,x-diol amounts. A procedure based on genetic algorithms was used to design simple neural networks. Two main goals are pointed out in this article: developing a general methodology for modeling the complex processes with simple or stacked neural networks, and demonstrating the improvement of the modeling performance by combining different neural networks.