Duplicate Detector
Paste real messages and watch the near-duplicate engine cluster them live, in your browser, with the similarity scores shown.
2 min · free · no account
Production thresholds are calibrated per deployment and run across every language at once.
Messages screened
8
Clusters found
1
Rows held back
3
Verdicts
Absolutely fantastic stay, the staff were amazing and the rooms were perfect!
Quarantined
1.000
Absolutely fantastic stay, the staff were amazing and the rooms were perfect
Quarantined
1.000
Truly fantastic stay. Staff were amazing, rooms were perfect!
Suspect
0.793
I waited three hours for a result nobody called me about.
Clean
0.221
The night pharmacy was closed again on Sunday evening.
Clean
0.212
Amazing stay! The staff were absolutely fantastic and rooms perfect!!
Quarantined
0.905
Nobody explained what the scan was going to cost me.
Clean
0.221
Best hotel ever, staff amazing, rooms perfect, absolutely fantastic!
Clean
0.706
1 coordinated cluster detected
In production these are held out of every analytic, kept fully readable, and shown with the exact signal that caught them. Removing one permanently takes an administrator and a typed confirmation.
This runs entirely in your browser, so nothing you paste ever leaves the page. Production screening works on meaning rather than characters, which additionally matches the same complaint across different languages.
Clusters above the threshold are the shape that catches coordinated praise waves: near-identical content arriving from distinct devices inside a short window.
Taxonomy Builder
Start from a sector template, edit every category, set escalation floors, and export the taxonomy your institution signs off.
OpenConfidence Calculator
Wilson 95% intervals on any share, with the sample size you would need to halve the uncertainty.
OpenLanguage Planner
Set your population's language mix and see exactly what share of it your current channel reaches.
OpenThese tools use the same logic
that runs in production.
Start a trial and point them at your own corpus instead of a sample.