clipping
Voter Grid Studio
server mode — SQLite + server-side Surya calls
Theme Dark
Pipeline Idle

🔒 Admin Access Required

Setup configurations and AI classifiers are locked.
Enter password to unlock this page.

Surya OCR endpoint

Requests go Python→Surya server-to-server — CORS never applies.

Grid detection & performance

Bulk Download (ECI Portal)

Generate ECI portal URLs to queue automated background downloads.

Idle

PDFs

Upload PDFs here, or batch-select rows below. Metadata extracted from page 1 during processing.

Pick one or more PDF files
idle
0 selected
File name Size Uploaded Status Action
Rows per page: page 1

Demographics Classifier

2-step system: first run the instant Rule-Based classifier (~70% coverage), then run the LLM classifier for the remaining ambiguous names. Manage and monitor everything here.

1
Fast Rule-Based Classifier

Pure Python surname lookup — covers ~70% of records instantly with no API calls. Run this first.

2
LLM Classifier (Ambiguous Names)

Sends remaining unclassified names to local Qwen-30B vLLM with 6 concurrent workers. Run after Step 1.

3
Jev Classifier (TypeSafe AI)

High-speed System One AI classification. Uses English transliterated names + Village demographics for high accuracy.

Declared classification data

Import caste/category values supplied by an authorized source. Records are matched only by EPIC ID; names are never analysed.

Download CSV template

Accepted columns: epic_id, caste_declared and category_declared. Category may be blank when a normalization rule supplies it.

Clean System 🔒

Wipe data to start fresh. Requires Admin Password.