How does Adam optimizer work internally?

Adam combines momentum and adaptive learning rates using moving averages.

It maintains first and second moment estimates of gradients, bias-corrects them, and updates parameters adaptively.