Build with us · open idea, needs builders

AI for road safety, built cheaper

The AI tools that make roads safer mostly exist already, but they are expensive. This page is about building the same nine jobs for a fraction of the cost, so ordinary people and small operators in India can afford them.

~1.7 lakhpeople die on Indian roads every year, the most of any country
9 ideas5 to start now, 4 for later
1 shared coreone phone app and one data pipeline serve 6 of the 9

Short answer: start with a crash database built from news (idea 1). Then build one phone app that holds the Hinglish driving coach and crash SOS (ideas 2 and 3).

"Someone already does it" is not a reason to skip an idea. The difference here is cost: the phone replaces special hardware, small models run on the phone instead of in the cloud, open data replaces paid data, and a self-serve app replaces a sales team.

Where the money goes

Why today's tools are expensive, and how to cut the cost

What costs money todayLow-cost replacement
Dedicated AI dashcams and sensor boxes in every vehicleThe phone people already own, or an old Android phone mounted in the vehicle
Uploading video to the cloud and processing it thereSmall models running on the phone. Only tiny results, like a pothole's location or a crash alert, leave it
Paid AI servicesFree hosted open models for the few text steps
Paid maps and street imageryOpenStreetMap, Mapillary open street imagery, public news
Survey vans and manual road auditsCrowd collection by volunteers, delivery riders and commuters using the app
Sales teams, pilots and contractsA self-serve app on the Play Store

The trick

One phone, many ideas

The phone already has a camera, motion sensors and GPS. Build the app that reads them once, and most of the ideas become small add-ons.

cameramotionsensorsGPS The phone Shared core small models on the phone+ one data pipeline+ open maps 2 · Driving coach in Hinglish3 · Crash SOS for budget phones 4 · Pothole map5 · Drowsy-driver alerts 8 · Danger-spot prediction9 · Pay-how-you-drive score
Mint: start now. Butter: later. Idea 1, the crash database from news, runs on its own pipeline and feeds idea 8.

The list

The nine ideas

1 · Crash database from news

Start now

TodayThe government publishes totals once a year. Detailed crash records are not open.

Low-cost versionAI reads local-language news every day and pulls out the place, time, vehicles and cause, building an open, street-level map. Free models and news feeds keep the cost near zero.

Who could fund or payGrants, CSR funds, researchers, journalists.

2 · Phone driving coach in Hinglish

Start now

TodayTelematics firms sell to insurers. Consumer apps are foreign and paid.

Low-cost versionThe phone's motion sensors score each drive on the phone itself and give tips in Hinglish.

Who could fund or payThe driver or a parent, through a small subscription.

3 · Crash detection and SOS for budget phones

Start now

TodayBuilt into Pixel and iPhone only, which are expensive phones.

Low-cost versionAn app for budget Android phones and two-wheeler riders that sends the location to family on WhatsApp or SMS.

Who could fund or payBundled with the driving coach.

4 · Pothole and road-damage map

Start now

TodaySurvey vehicles and fees charged per kilometre.

Low-cost versionA phone on the dashboard runs a small detection model, and commuters and delivery riders build an open map.

Who could fund or payGrants first, cities later.

5 · Drowsy and distracted driver alerts

Start now

TodayAI dashcam hardware plus a monthly fee per vehicle.

Low-cost versionAn old phone faces the driver and checks for closed eyes and phone use on the device, with nothing sent to the cloud.

Who could fund or paySmall fleets with 1–10 trucks.

6 · Night voice companion for truck drivers

Later

TodayLittle exists.

Low-cost versionVoice chat in Hindi that checks the driver is still alert at night.

Who could fund or payThe fleet or the driver's family.

7 · Helmet and triple-riding detection

Later

TodayBig city camera contracts.

Low-cost versionA cheap camera with a small edge box, or existing CCTV feeds.

Who could fund or payFactory gates, campuses and housing societies.

8 · Danger-spot prediction

Later

TodayThe iRASTE pilot in Nagpur, plus manual road-safety audits.

Low-cost versionBuilt from idea 1's crash data, idea 4's pothole map and OpenStreetMap roads.

Who could fund or payHighway authorities and cities.

9 · Pay-how-you-drive insurance score

Later

TodayInsurer telematics programmes using a device or app contracts.

Low-cost versionReuse idea 2's driving scores as a report the driver shares with their insurer.

Who could fund or payInsurers, once idea 2 has users.

Order

Order of work

  1. Idea 1, crash database. It is the cheapest, a public good and grant-friendly, and it creates the data the other ideas need.
  2. Ideas 2 and 3 as one phone app. This builds the shared phone core.
  3. Ideas 4 and 5 as add-ons to the same app.
  4. Ideas 6 to 9 once there are data and users.

First small step for idea 1: pick one state, list 20 Hindi and English local news feeds, extract one week of crash reports with a free model, and count how many give a usable place name.

Join

Want to build one of these?

All nine are open. Each one needs a lead, and idea 1 is a good first project for a data engineer who wants to learn language models on real, messy data. Write to basant@cleandataengineering.ai with the idea number you want to pick up.

Sources, from prior knowledge and to be rechecked before building: Ministry of Road Transport and Highways, "Road Accidents in India" yearly reports; iRASTE Nagpur (Intel, IIIT Hyderabad); crash detection on Pixel and iPhone; IRDAI's 2022 approval of telematics add-ons.