You need to implement a CSV parser which has the capable of parsing group of record using a pre-defined mapping file. The mapping file is the source of identifying CVS records pattern and you need to emit de-serialized objects containing parsed values of CSV file.

Example for a record group:-

header,12245,25-10-2012

body,this is a test value-1,12254,1

header,12544,31-12-2014

body,this is a test value-2,25441,4

http://ruby-doc.org/stdlib-1.9.2/libdoc/csv/rdoc/CSV.html

https://docs.python.org/2/library/csv.html