Raaste

Parking-congestion intelligence and enforcement planning for the Bengaluru Traffic Police.

Wrong and illegal parking is one of the biggest sources of street-level congestion in Bengaluru, but it's hard to see in the data: every ticket is just a dot, and a dot doesn't tell you which corners actually choke the city or what they cost. Raaste turns the enforcement record into something a traffic officer can act on — where the problem is, what it's worth, where congestion is going untouched, and what to patrol tomorrow.

What it does

The app is one map-first dashboard plus two operational views, mirroring the chain of people who'd actually use it.

• / — City dashboard. A live hotspot map (heatmap + ranked corners) you can scrub by hour of day and day of week. On top of it:
  • an impact score that weights each violation by how much it obstructs traffic, so a blocked main road outranks a quiet side street;
  • a congestion overlay that lines the parking hotspots up against real ASTraM congestion/incident events — the proof that these corners and the city's congestion are the same places;
  • an enforcement blind-spot lens: corners with heavy logged congestion but almost no parking enforcement — where attention should go and isn't;
  • a forecast that re-ranks corners by predicted violation intensity for any hour/day, and an optimizer that splits the top corners into patrol routes;
  • supporting panels for the recoverable fine value, a what-if enforcement simulator, and a per-junction drilldown.
• /briefing — Patrol beat sheet. Pick a station and a shift and get a printable one-pager: the corners to cover in driving order, expected violations, peak window, the dominant offence and vehicle, nearby congestion, and the fine value sitting there if enforced.
• /field — Field mode. The beat sheet, made for a constable's phone. Hand it off from the beat sheet with a QR scan, then work one corner at a time: live distance to the next corner, a tap to open native maps, log each stop as issued / cleared / skipped, watch the running tally, and finish with a printable shift report. Stays usable when GPS is off and resumes mid-shift if the page is reloaded.

Data

Built on two anonymized, geocoded datasets:

• ~298,000 Bengaluru Traffic Police parking-violation records, November 2023 – April 2024 (location, vehicle type, offence, timestamps, police station, junction).
• ASTraM congestion and incident events (congestion, accidents, breakdowns, water-logging) with priority, road-closure and corridor/junction context, used for the congestion overlay and the blind-spot analysis.

The raw exports are large and aren't committed. A small Python pipeline reads them and writes compact JSON aggregates into public/data/, which is what the app loads — so the site runs without the raw data.

Tech

• Next.js (App Router, TypeScript) and React, deployed on Vercel.
• Map: the basemap is MapmyIndia / Mappls, with deck.gl drawing the hotspot, congestion and blind-spot layers on top. When a Mappls key isn't present (local dev, preview URLs) it falls back to CARTO raster tiles via MapLibre GL, so the map always renders. Tailwind CSS for styling; charts are hand-built SVG, no chart library.
• Python for the pipeline — standard library for the data shaping, numpy + scikit-learn for the model. The forecast is a random-forest regressor predicting violations per ~220 m cell per hour-of-week (R² ≈ 0.65 on a held-out test split).

Running locally

npm install
npm run dev

Then open http://localhost:3000. The committed aggregates in public/data/ are enough to run everything. Locally the map uses a CARTO basemap; set NEXT_PUBLIC_MAPPLS_KEY (a domain-whitelisted Mappls web key) to render the MapmyIndia basemap in production.

Regenerating the data (optional)

Only needed if you have the raw CSVs and want to rebuild the aggregates.

python -m venv .venv && source .venv/bin/activate
pip install -r pipeline/requirements.txt

# point the scripts at the violations CSV (or drop the files in ./data/raw/)
export RAASTE_RAW=/path/to/violations.csv

# each script writes one or more files into public/data/
python pipeline/build_data.py
python pipeline/build_congestion.py
python pipeline/build_junctions.py
python pipeline/build_fines.py
python pipeline/build_simulator.py
python pipeline/build_blindspots.py
python pipeline/train_model.py

Layout

src/app/            routes — / (dashboard), /briefing (beat sheet), /field (field mode)
src/components/     map, panels, and the field-mode workflow
src/lib/            beat-sheet logic, routing/geo, shift log, formatting, types
pipeline/           Python scripts: raw CSV -> public/data/*.json
public/data/        committed JSON aggregates the app reads

Credits

Built for the Bengaluru Traffic Police parking-congestion challenge (Gridlock 2.0). Mapping by MapmyIndia / Mappls; the enforcement and congestion datasets come from the Bengaluru Traffic Police and its ASTraM programme.