will segments be built per input file for spark based batch ingestion or does data get repartitioned to ensure segments are not tiny?

The former I htink.

any compaction i can do with in pinot? or has to be done before ingesting into pinot?

You can checkout merge/rollup tasks.