I'm not sure whether your question concerns what mathematical transformations are used or what the geodetic reference parameters are for a particular projection. Perhaps you are concerned about both issues.
To learn about the geodetic reference parameters are for a particular projection, refer to GeoProjectionData.
To learn about the mathematics of a particular projection, Google on the projections name. To learn what projections are available, again refer to GeoProjectionData.
However, one can make a good guess about the mathematical transform directly from the result given by GeoGridPosition by posing a request that uses distinctive primes as latitude and longitude values.
For example:
GeoGridPosition[GeoPosition[{17, 43}], "Mercator"]
GeoGridPosition[{43, (180 Log[Cot[(73 π)/360]])/π}, "Mercator"]
Noting that 73 is the colatitude of 17, this suggests that the transform is
toMercator[lat_, long_] := {long, 180 Log[Cot[(90 - lat) π/360]]/π}
Some tests to confirm this finding.
toMercator[17, 43]
{43, (180 Log[Cot[(73 π)/360]])/π}
toMercator[45, -45]
{-45, (180 Log[Cot[π/8]])/π}
This works for other projections as well. Here is a more complex but still informative example.
GeoGridPosition[GeoPosition[{17, 43}], "Albers"]
GeoGridPosition[
{4 Sqrt[1/3 (1-1/2 Sqrt[3] Sin[(17 π)/180])] Sin[(43 π)/(240 Sqrt[3])],
4/Sqrt[3]-4 Cos[(43 π)/(240 Sqrt[3])] Sqrt[1/3 (1-1/2 Sqrt[3] Sin[(17 π)/180])]},
"Albers"]
I won't work out a Mathematica projection function from this, but I think you can see the form it would take fairly well.