Shital InfotechElectoral Roll Extraction Pipeline — government data at scale
GPU-OCR turns millions of scanned voter roll PDFs into a verified, structured dataset.
Turns millions of scanned, Hindi-language government voter records into a clean, ready-to-use spreadsheet — reading each record with AI, then double-checking every single one against the official government database.
The problem
Electoral rolls are only published as scanned image PDFs — there's no downloadable dataset. Off-the-shelf OCR tools produce garbled, unusable text at this scale, so getting a clean, trustworthy dataset out of them takes a purpose-built pipeline.
How we built it
Every record is read from the scanned pages using AI-powered text recognition tuned specifically for this document layout and script, then every single one is independently verified against the government's own official portal — so the final dataset isn't just extracted, it's confirmed accurate.
What it does
- Every record cross-checked against the official government portal — not just OCR guesswork, genuinely verified data
- Built to handle millions of records across multiple districts without manual review
- Automatically works around the government portal's bot protection so verification doesn't get blocked mid-run
- Clear source tagging on every record — you always know whether a field came from the scan or the verified portal
- Designed to run cost-effectively on rented cloud GPU capacity rather than dedicated hardware
- Delivered as a clean spreadsheet plus a run summary — total records, verified count, and any records worth a second look