Ward-level safety map for Indian cities. 43 cities across 28 states + 6 UTs, in 13 languages. Per-ward risk scores derived from real city-aggregate crime statistics (NCRB / state-police annual reports), with a night-time multiplier and a women's-safety filter. Five cities (Mumbai, Delhi, Kolkata, Pune, Gurugram) also republish the official police Absconder / Proclaimed Offender list with attribution.
Live at https://eeman1113.github.io/crimeradar/.
Per-area risk shaded against a real city-aggregate total, with night-mode and women's-safety toggles, a "use my location" jump, and a side panel listing the highest-risk wards.
Mumbai · dark theme · BMC 24 wards · calibrated to YTD Jan–Mar 2026 |
 Gurugram · light theme · MCG ward heatmap · 2022 annual |
Drill into any ward for the YTD category breakdown, the per-1,000-resident rate, and recent crime-related news headlines (last 18 months, Google News RSS, keyword-filtered).
All 43 cities side-by-side. Bar chart normalises to cases per 100,000 residents so big metros don't drown out smaller capitals; the radar chart shows the shape of crime (theft-heavy vs assault-heavy vs sexual-offence-heavy) regardless of volume.
 Per-100k rates across all 43 cities, each year-window stamped |
 Crime-mix radar — each category normalised to the highest city = 100 |
City names, UI strings, and number formats localise to any of 13 Indian languages. City names come from Wikipedia interlanguage links; UI from a hand-maintained dictionary.
 Language picker — English + 12 Indic locales |
 Home page in বাংলা — city tiles re-render in script |
- Ward boundaries — sourced per-city from datameet / datta07 / ESRI India Living Atlas / OSM Overpass / official municipal GIS portals.
- City-wide crime totals (43 cities) — NCRB "Crime in India 2022"
(megacities + District-wise Additional Tables), NCRB 2023 expanded
34-city table, plus state-police annual reports / commissionerate PDFs
for capitals. See
data/cities/<id>/monthly_stats.jsonfor each city's source URL + year. - Per-ward populations (26 of 43 cities) — Census 2011 (or the city's own published voter/ward demographics where they predate Census). The remaining 17 cities have post-2011 ward delimitations that don't map 1-to-1 to Census ward IDs; they use an even-split estimate over the total city population until a spatial join is built.
- Per-ward risk distribution — apportioned from each city's real
city-wide totals using a tier model (central, inner, outer, peripheral
bands derived from each ward's distance to the city centroid). The
calibrate()step inlib/wards.tsthen rescales every ward so the category sums match the real city number. The data-quality badge on each city card reads "calibrated" when this real-sum match is in effect. - News headlines per ward — Google News RSS, keyword-filtered for crime/police/arrest terms, ≤4 items per ward, refreshed weekly.
- Localized city names (12 Indic languages) — Wikipedia interlanguage links. Average coverage 11/12 across 43 cities.
Next.js 16 (App Router) static export · TypeScript · Tailwind v4 · MapLibre GL JS · Recharts · GitHub Pages.
The city registry is manifest-driven — data/cities.manifest.json is the
single source of truth. scripts/codegen_cities.mjs runs as a prebuild
hook and emits three generated files (lib/cities.generated.ts,
lib/wards.generated.ts, lib/absconders.generated.ts). Adding a city is
one manifest row + a GeoJSON file in public/geo/.
Per-city data lives under data/cities/<id>/:
data/cities/mumbai/
wards-raw.ts per-ward seed (id, name, population, breakdown)
monthly_stats.json city-wide annual/YTD totals from the official source
monthly_stats_history.json multi-year history where available
ward_news.json Google News RSS items per ward
absconders.json police absconder list (5 cities only)
# Drop the ward GeoJSON in public/geo/<id>_wards.geojson, then:
node scripts/add_city.mjs \
--id <id> --name "<Name>" --state "<State>" --state-code IN-XX \
--tier capital --geojson public/geo/<id>_wards.geojson \
--ward-id-key <propertyName> --unit "<MC ward>" \
--population <city_total> \
[--population-key <perWardPopProperty>] # if GeoJSON has itThe script validates the GeoJSON, registers the city in the manifest, stubs the four data JSON files, generates the ward seed, and re-runs codegen.
npm install
npm run dev # auto-codegens via prebuild, then next devOpen http://localhost:3000.
Three GitHub Actions workflows keep the data fresh:
| Workflow | Cadence | What |
|---|---|---|
ingest_news.yml |
Sunday 04:00 UTC | Google News RSS per ward for every city in the manifest |
ingest_stats.yml |
1st of month 03:00 UTC | Mumbai / Bangalore / Chennai / Delhi monthly scrapers + optional NCRB CSV ingest via workflow_dispatch |
ingest.yml |
Monday 02:30 UTC | Absconder lists (5 cities) |
deploy.yml |
Every push to main |
next build static export → GitHub Pages |
Manual trigger:
gh workflow run ingest_news.yml
gh workflow run ingest_stats.yml -f ncrb_csv_url=https://... -f ncrb_year=2023# Once you have the megacity table as CSV (City, Crime Head, Cases):
node scripts/ingest_ncrb.mjs --csv ncrb_megacities.csv --year 2022Aliases for state-capital names are wired into the same pipeline, so state-police CSVs in the same shape ingest the same way.
13 locales supported (en + 12 Indic). UI strings live in
lib/i18n/strings.ts; city names come from nameI18n per city in the
manifest (populated via scripts/fetch_city_names.mjs from Wikipedia
langlinks). useI18n() exposes both t(key) and cityName(id).
Risk scores are estimates, not safety guarantees. See /methodology
for the exact formula, every data source, and known limitations per
city. See /legal for the naming policy (absconders are only on this
site because the police themselves published the names), takedown
contact, and DPDP notice.
This project is licensed under the GNU Affero General Public License v3.0 — see LICENSE.