Regarding mapping of anusvara .

[arushidogra]

The anusvara is always mapped to म् which is wrong as it is sometimes mapped to न् and sometimes to म् depending on the succeeding alphabet

Here the output should have been chandan.

Here the output should have been sambandh

[copyninja]

I agree there is variation in the pronunciation and definitely this is not easy to be represented in English. But if you have approach to deduce this programmatically feel free to propose a patch.

[stultus]

I think If the letter following anusvara is of the Alveolo-palatal ('ta' family) and Dental ('tha' family) consonant, we can use 'n' instead of 'm'.