Model prediction uncertainty

Looking for a consultant to help us understand methods for determining the uncertainty associated with models we use in the semiconductor industry. Our models are multivariate and nonlinear, but they are trained on input data collected from semiconductor fabs, and for each of thousands of samples (with different characteristics) measured, there are multiple repeat measurements done across some spatial domain such that the mean values for each characteristic are entered into the model training, but in some cases the 95% CI associated with determination of each mean value is also known. We are trying to understand generally how to determine the MODEL PREDICTION 95% CI for the two scenarios of a) only mean values input to the model and b) both mean and 95% CI for measured input data are available. It seems to us that for simple first order polynominal regression (y = mx+b) a model prediction uncertainty can be calculated for each of the above scenarios, is that true? And how to do it? Then, for our much more complicated multivariate nonlinear models, does the determination of model prediction uncertainty become untenable or are there known methods to nonetheless calculate or estimate this uncertainty associated with the predictions. So we are seeking a consultant who can educate us on the general concepts associated with model PREDICTION uncertainty (like distinguishing between model confidence interval in predicting something like mean and model prediction intervals for any individual measurement..) Then with a description of our much more complex models, guide whether such model uncertainty assessments are quantitatively possible.