Built to fix a real encoding problem, not just a demo.
The Election Commission's own PDFs break Telugu text search: their Type3 font encoding scrambles the same name into different byte sequences depending on when it was typed in. ManaVoter ingested those PDFs with Tabula-java and Apache PDFBox, then ran every name through a consonant-skeleton normalization algorithm I designed and built myself, stripping diacritics down to canonical forms so variant spellings still matched.
The result indexed 1.23 crore voter records across six districts, served from a thread-safe immutable in-memory index built with Spring Boot's @PostConstruct, so concurrent reads never hit synchronization overhead. That's what kept lookups under 100ms.
It ran in production on Railway with a custom domain and SSL, picked up 300+ users and 2,300+ tracked events with no marketing at all. The pilot proved the approach and found its users. That phase is done, and the live search is currently offline.