On-device AI
Detection runs locally through Core ML. The default model is simple: camera in, structured parking signal out.
SIG · OVERVIEW
On-device computer vision
Parking,
found fast.
ranked by freshness.
shared as signal.
clear enough to act.
Peer2Park turns a passing phone camera into a lightweight parking signal. Detection happens on device, the freshest report rises first, and raw video stays local by default.
scroll
18ms
On-device inference
3
Signal stages
8
H3 resolution
0
Raw video upload
On-device detection
Structured signals only
Freshness-ranked routing
No raw video upload
H3 cell grouping
Driver confirmation loop
Privacy-first design
Lightweight cloud path
On-device detection
Structured signals only
Freshness-ranked routing
No raw video upload
H3 cell grouping
Driver confirmation loop
Signal flow
How a parking signal
moves.
This panel is an illustrative walkthrough of the product flow, not a live network view. It shows the product logic in the same visual language as the rest of the site.
Shared fields
0
core
Freshness window
0
min
Dedup radius
0
m
Median inference
0
ms
peer2park · example-flow.log
Example
Status quo
Parking apps show you where spots were.
- ×Static maps show general supply, not actual curb availability.
- ×Even a small delay can turn a reported spot into a wasted turn.
- ×Drivers still guess whether a spot will be open when they arrive.
- ×Drivers circle blocks, wasting time, fuel, and patience.
Peer2Park
Freshness over prediction.
- ✓On-device vision detects openings from a passing camera view.
- ✓Signals can be confirmed, re-ranked, or aged out as conditions change.
- ✓Results are ordered by recency and confidence.
- ✓H3 indexing keeps nearby detections grouped and deduplicated.
The Solution
Built on freshness, not predictions.
Freshness-ranked
A newer, stronger signal wins. Older reports decay instead of lingering like static map data.
Crowdsourced mesh
Multiple drivers can reinforce, invalidate, or replace the same curb event as conditions change.
Detection demo
See the detection layer,
not just the map.
The clip below is a product demo of the curb view the model reads. The point is faster decisions without sending raw video to the cloud.
Video failed to play in this preview. Open the file directly or switch preview browser.
DEMO · CV
How it works
From passing frame to
useful signal.
01
Capture
A driver passes the curb
The phone camera observes the street during a normal trip. No extra hardware or dedicated scan route required.
02
Detect
The model detects an opening
On-device inference estimates whether the space is open and how confident the resulting signal should be.
03
Route
Nearby drivers see the newest signal first
The result is grouped, deduplicated, and ranked by freshness before it is surfaced.
Under the hood
Technical where it matters,
minimal where it doesn't.
Peer2Park combines on-device detection, geospatial grouping, and lightweight cloud routing. Enough infrastructure to be useful, without turning the product into a surveillance system.
Core ML
YOLO11
H3
Apple Maps
Serverless
Structured signals
Privacy-first
H3 hexagonal indexing
Nearby detections collapse into consistent cells so the app can route and deduplicate quickly.
Intelligent navigation
The product can hand a fresh destination into the mapping layer drivers already use.
Privacy-first by design
The shared object is metadata, not camera footage. The default flow keeps raw video on the device.
Serverless backend
A lightweight backend can score freshness, suppress duplicates, and deliver nearby results.
Why freshness matters
Parking data expires
fast.
A report that is seconds old can still help. A report that is several minutes old can already be wrong. Peer2Park treats time as part of the product, not a footnote.
Signals carry a timestamp, a confidence score, and a decay window so stale results fall away instead of cluttering the map.
Elapsed 0s
The curb is a network. Every driver is a sensor. Parking is a signal.
Peer2Park is not a promise that a spot will still be there. It is a faster, cleaner way to act on the best signal available.
Start a real conversation
Pilots, campuses, fleets,
and city use cases welcome.
If you want to test the product, review the data model, or discuss deployment constraints, reach out.