Skip to content
← The Workbook

No. 31 · 2026Concept

Plate Karma

Yelp for drivers. A Jetson on the dashboard, camera scanning license plates, 15-minute rolling video buffer. Cut me off, miss your blinker, brake-check the family minivan? Press the button, the incident saves with the plate. Plates cross-reference across the network. Repeat offenders surface. At scale, it's community-sourced asshole detection on the road.

Status
Concept
Year
2026
Stack
Jetson · Computer Vision · Networked · Community · Driving

The Idea

Yelp for drivers. The plate is the username. Anonymous to the human, identifiable to the system.

A Jetson Orin Nano sits on the dashboard with a small camera. It runs license-plate recognition on the live feed and keeps a rolling 15-minute video buffer. Most of what gets seen gets thrown away. When something noteworthy happens, somebody cuts me off, runs a red, brake-checks the minivan in front of them, I press a button. The buffer locks, the plate gets pulled, the incident gets logged.

How It Works

Hardware: Jetson Orin Nano, dashcam-grade camera, dash-mount, single physical button wired to GPIO.

Software: ANPR runs on every frame, rolling video buffer in memory, button press writes the previous N seconds plus the next N seconds to disk with the OCR'd plate as the filename. The record drops into SQLite: plate, timestamp, video path, behavior tag, free-text note.

That's the local loop. The interesting part is the network.

The Yelp Layer

One person's incident is anecdote. A hundred people on the same plate is a pattern.

Each device shares plate-level events to a community database. No identity, no driver name, no DMV lookup. Just plate plus behavior plus frequency. If a plate gets ten reports in a month, repeat offender. If it gets a hundred, the database has enough signal to do something with.

Two outputs at scale:

  1. Real-time warning: nearby flagged plates get surfaced to the user the way Waze surfaces speed traps. The roads get a shared memory.
  2. Report packet: at threshold, the system compiles video evidence plus cross-reference timeline plus aggregate behavior pattern, and routes it to whichever law-enforcement portal accepts community submissions in that jurisdiction.

What Could Go Wrong

Plates are PII. Crowdsourced behavior reports on a person are a defamation surface. Harassment vector. False reports could ruin a stranger's day. The Yelp metaphor has the same problem Yelp has, except the target can't reply.

Mitigations to think through, not solved yet:

  • Threshold gating before any plate is publicly visible (10? 100? still arbitrary)
  • No driver-identity lookup, ever (plate only, no DMV cross-ref)
  • Required video evidence on every report (no text-only complaints)
  • Time-decay so a single bad day fades out of the record
  • Some kind of dispute path for the wrongly-tagged

Doesn't make the privacy problem go away. Worth being honest about.

Status

Concept. Hardware on hand from the Red Phone family. Proof-of-concept pile.