Electoral intelligence for India at booth level — built party-neutral, and built so that every number on screen can be traced back to the source it came from.
Political data in India is abundant, scattered and mostly unusable: results sit in one archive, boundaries in another, demographics in a third, and none of them agree on what a constituency is called. The hard part was never the dashboard. It was assembling a trustworthy base out of public record — and being able to prove, row by row, where every figure came from.
Every ingested record keeps the lowest granularity available and rolls up from there — booth to assembly constituency to parliamentary constituency to district to state. Aggregate first and the detail is gone for good; keep the atom and every view above it is a query. Constituency matching, name normalisation and party normalisation run as their own stage, because the same seat is spelled four ways across four decades of records reaching back to 1951.
The DPDP Act and the Election Commission's Model Code are constraints in the schema, not paragraphs in a document. Every row carries its source, fetch URL, timestamp and licence tag. Sensitive categories are encrypted at rest and gated behind a second permission check. Reads are written to an append-only audit log. And the rule above all of them: no capability may depend on which party is asking — ruling and opposition get the identical product.
Ingestion is declarative rather than bespoke: a registry names every source with its licence, cadence and owner, and eleven loaders pull against it — historical results, boundary maps, electoral rolls, fact-check feeds, government wires. Authoritative beats scraped every time; a scrape has to justify itself by the absence of an official channel.