LokDarpan
A public-finance and governance intelligence platform for India — links official records into one traceable, auditable ledger from revenue to completed work, with compiler-enforced neutrality and provenance. My most ambitious build to date.
“Every rupee traceable, every observation neutral — a mathematical-consistency engine over India's public finance and infrastructure records.”
Role in the platform
The flagship. A standalone, public-interest platform — independent of the AI-developer product family below — and the most demanding system I've designed: in scope (a full revenue-to-completion ledger across 99 government sources), rigor (110 specification documents written before implementation), and the neutrality/provenance discipline the domain requires.
Overview
LokDarpan is a public-finance, governance, and infrastructure intelligence platform for India, built entirely on official government records. It links official data into one traceable ledger — revenue → budget → ministry → state → district → local body → department → scheme → tender → contractor → release → expenditure → work progress → completion → audit — and runs mathematical-consistency and variance checks over the chain. It is my most ambitious build to date: not in lines shipped yet, but in scope, rigor, and the discipline the domain demands.
It is explicitly not an anti-corruption or accusation engine, a legal authority, or a source of allegations. It is a transparency tool, a mathematical-consistency checker over official records, and an anomaly highlighter with full source traceability — every figure is traceable to an official source, and every observation is a neutral, factual statement.
The problem
Public-finance and infrastructure data in India is scattered across dozens of government portals, in incompatible formats, with no way to trace a rupee from the budget line that allocated it to the completed (or incomplete) work on the ground.
Existing dashboards summarize a single source in isolation; nothing links revenue → budget → scheme → tender → expenditure → completion into one auditable chain, or checks whether the numbers along that chain are internally consistent.
Any tool in this space has to be provably neutral: a mislabeled "anomaly" can read as an accusation, so the platform's credibility depends on discipline most dashboards never have to build.
What it is
A pnpm/Turborepo-style monorepo: apps/web (the Next.js public site, the first product), services/ (ingestion, normalization, entity-resolution, analytics, a risk engine, an AI layer, and the public API), packages/ (money for exact-precision ledger arithmetic, neutrality for a compile-time-enforced neutral-language gate, contracts, domain, config, database, observability, errors), database/ (versioned, immutable migrations against Postgres + PostGIS), and .docs/ — 110 specification documents written before implementation began.
Money is stored as bigint paise, never a float, because a national multi-year aggregate exceeds Number.MAX_SAFE_INTEGER; a <Figure> component cannot render a number without a provenance prop; and observation text is typed as ServerText so neutral copy can only originate from the API, with a CI language gate (pnpm neutrality) blocking release on a forbidden-language hit.
Key features
Tech stack
Why it's different
- Built on a traceable ledger chain, not isolated dashboards — every figure connects back through the full revenue-to-completion path.
- Specification-first discipline: 110 documents and binding legal/ethical rules were written before a line of ingestion code, because a factual error at this scope is a credibility failure, not a bug.
- Neutrality and provenance enforced by the compiler, not by policy — the type system and a CI language gate make the platform's core promise ("every figure sourced, every observation neutral") structurally true.
- Deliberately scoped as a public-interest project, independent of any government body, agency, or political party.