@aialok/lakhua

Sub-millisecond reverse geocoding for India, fully offline.

No API calls. No network latency. No rate limits.

Built for backend services, cron jobs, and analytics workloads that need deterministic geocode lookups.

Features

• 📍 converts lat, lon to city, state, optional district and pincode
• 🔢 supports direct H3 index lookup via geocodeH3()
• ↩️ parent-cell fallback (resolution 5 → 4) when exact cell has no data
• ⚡ data loaded once per process — all subsequent lookups are in-memory map reads
• 🐛 optional debug mode traces load time and per-lookup timing
• 🔷 TypeScript-first — full type definitions included

Why Not Hosted APIs?

| Feature | @aialok/lakhua | Hosted API | |---|---|---| | Works offline | yes | no | | API key required | no | usually yes | | Network dependency | no | yes | | Per-request cost | no | usually yes | | Latency variability | low, in-process | network-dependent |

Installation

npm install @aialok/lakhua
# or
bun add @aialok/lakhua

Quick Start

import { geocode } from "@aialok/lakhua";

const result = geocode(28.6139, 77.2090);
if (result) {
  console.log(result.city, result.state);
}

API

Top-level functions (recommended)

geocode(lat: number, lon: number, options?: GeocodeOptions): GeocodeResult | null
geocodeH3(h3Index: string, options?: GeocodeOptions): GeocodeResult | null

These use the internal singleton geocoder — no class instantiation needed.

GeocodeOptions

interface GeocodeOptions {
  resolution?: number;  // H3 resolution for geocode(lat, lon). Default: 5
  fallback?: boolean;   // Walk up to parent resolution on miss. Default: true
  debug?: boolean;      // Print load and lookup timings. Default: false
}

GeocodeResult

interface GeocodeResult {
  city: string;
  state: string;
  district?: string;
  pincode?: string;
  matched_h3: string;         // H3 cell that matched (may be parent)
  matched_resolution: number; // Resolution of the matched cell
}

Returns null for invalid input or when no data exists for the given location.

Advanced class APIs

import { ReverseGeocoder, ReverseGeoDataLoader } from "@aialok/lakhua";

ReverseGeocoder.getInstance()
ReverseGeoDataLoader.getInstance()

Use these only when you need explicit control — e.g. testing or custom singleton lifecycle.

Examples

Coordinate lookup

import { geocode } from "@aialok/lakhua";

const result = geocode(12.9716, 77.5946); // Bengaluru
if (result) {
  console.log(result.city, result.state, result.pincode);
}

Direct H3 lookup

import { geocodeH3 } from "@aialok/lakhua";

const result = geocodeH3("8560145bfffffff");
if (result) {
  console.log(result.city);
}

Debug mode

import { geocode } from "@aialok/lakhua";

const result = geocode(19.076, 72.8777, { debug: true });
// prints load + lookup timings to console

Disable fallback

import { geocode } from "@aialok/lakhua";

const result = geocode(28.6139, 77.2090, { fallback: false });
// only checks resolution 5, no parent lookup

How It Works

1. Convert lat, lon to H3 at resolution 5 (or configured resolution).
2. Lookup the H3 key in in-memory store.
3. If not found and fallback is enabled, check parent resolution 4.
4. Return a result object, or null if no match exists.

lat, lon -> h3(res5) -> in-memory map -> optional parent(res4) -> result/null

Data Source and Indexing

• Indexing system: Uber H3
• Geographic source data: OpenStreetMap data by OpenStreetMap contributors
• Distribution model: precomputed JSON stores bundled with the package

Data Coverage

• India-focused dataset
• H3 resolutions: 4 and 5
• Fields: city, state
• Optional fields: district, pincode (when available)

Runtime Compatibility

• Node.js
• Bun

@aialok/lakhua reads packaged JSON files from local filesystem at runtime.

Performance

• Data is loaded into memory once on first call.
• Each lookup is a single map read — typically < 1ms.
• With fallback enabled, up to 2 map reads (resolution 5, then 4).

Benchmarking

Use test/performance.test.ts to benchmark in your own environment. When publishing numbers, include machine type, runtime version, and warm/cold process context.

When Not To Use

• You need rooftop-level precision
• You need global coverage
• You need live POI or continuously updated location feeds

Development

bun install
bun run lint
bun run format
bun run typecheck
bun test

License

MIT