You haven't mentioned whether you are using Zebra.
To search successfully in Devanagari you will have to use Zebra.
You can apply http://git.biblibre.com/cgi-bin/gitweb.cgi?p=koha;a=commitdiff_plain;h=b4068c1da33de36019bf4192150b6234bb8c7dbf;hp=a739ed37ef5e289cf6ac9815acd9fc6ede052bfc this patch to make necessary changes to you Zebra Configuration..

Or if you are not comfortable applying patches etc you can follow following procedure..

#sudo apt-get install libicu-dev libicu38 yaz-icu

Copy following files from /usr/share/idzebra-2.0/tab/
to /etc/koha/zebradb/etc/
icu.idx
phrases-icu.xml
string.chr
words-icu.xml

Copy /etc/koha/zebradb/etc/words-icu.xml to /etc/koha/zebradb/etc/icu.xml

Edit /etc/koha/zebradb/etc/icu.xml change <icu_chain locale="en"> to <icu_chain locale="en_IN.UTF-8">

Save and change ownership of all copied files to koha:koha

#sudo chown -R koha:koha /etc/koha/zebradb/etc

Edit /etc/koha/zebradb/zebra-biblios.cfg
add line
index: icu.idx

Restart zebra server and queue daemons
#sudo /etc/init.d/koha-zebra-daemon restart
#sudo /etc/init.d/koha-zebraqueue-daemon restart

Rebuild zebra index
#sudo KOHA_CONF=/etc/koha/koha-conf.xml PERL5LIB=/usr/share/koha/lib /usr/share/koha/bin/migration_tools/rebuild_zebra.pl -b -r -v

Using yaz-client to verify zebra is working
#az-client
z>auth kohauser zebrastripes
z>open unix:/var/run/koha/zebradb/bibliosocket
z>base biblios
z>f ”पाँडे”
z>s
z>q


Adjusting Koha System Preferences to correctly search our multilingual data
In Koha staff client
Home → Administration → System Preferences → Searching
Turn off QueryFuzzy ( maybe and QueryWeightFields)

Regards,
Koustubha Kale
Anant Corporation

Contact Details :
Address  : 103, Armaan Residency, R. W Sawant Road, Nr. Golden Dyes Naka, Thane (w),
        Maharashtra, India, Pin : 400601.
TeleFax  : +91-22-21720108, +91-22-21720109
Mobile     : +919820715876
Website  : http://www.anantcorp.com
Blog : http://www.anantcorp.com/blog/?author=2