I believe, this is how it works. The sanskrit consonants (excepting the
"ya ra la va ...") can be viewed as a matrix with each row indicating
the vargam (ka-vargam, cha-vargam,...) and each column similarly corresponding
to the i'th character.
Then, the last column is called the "anu-nasikams" (each row and column have
names, the rows depending where your tongue touches the part of the mouth,
e.g., dantamulu, ...). The notion of putting consonants next together with
"zero" (sunna) is a violation of sanskrit's pure phonetic rules. Indeed,
"zero" is an approximation for the anunasikam that goes there.
The proper anunasikam is determined by the second consonant that goes with the
first one.
Thus, gampa is gampa, but not ganpa or gaNpa because "ma" is anunasikam of
pavargam (pa, pha, ba bha, ma).
Simalarly, kuNda is KuNda, not Kumda, and gaNta, not gamta, and similarly
the others Pillala Marri pointed out above
Bala Subrahmanyam makes mistakes like this. The most obvious one was
in the song describing Arjuna's fight with shiva in which he says "pamdi".
It should be "pandi" (ta, tha, da dha na). Somehow he always throws in
"m" when the right thing is not "m".
On a similar note, observe where your tongue touches the mouth when you
say ka-vargam, cha-vargam, ta-vargam, and pa-vargam. It touches exactly
the same place for any given vargam (only difference being voiced or not,
and aspirated or not -- making the four consonants, and throw in an appropriate
nasal (anunasikam) and you get the entire vargam).
Sanskrit is a linguist's delight.
Kumar