We present an efficient and consistent method for discerning preferable and generating functions of data using filtered spectral fingerprints. Spectral analysis of functions and data is hardly new. For example, with Fourier Transforms one can classify and identify functions using coefficients. The problem is that statistical, domain and systematic error noise is baked into the spectrum of coefficients. For over two centuries Legendre polynomials have been used for spectra, however, we are unaware of prior use of this approach. We create a hash library of function spectra to quickly find preferable and/or generating functions. Software tools (Patents Pending) have been developed utilizing this methodology, including one for finding AI/ML Neuron Activation functions. Even for noisy and instrument biased data, researchers and machine optimizers can better discern preferential and underlying functions that generate noisy data. Preliminary AI/ML results are also presented.