Five-model ensemble
Five models evaluate the same completed interaction, reducing reliance on any single model result.
How Ramon works
Quiet during the survey. Clear after it.
Ramon never blocks, steers, or classifies respondents in real time. It evaluates the completed interaction after the survey ends.
- 01
Collect signals.
A small JavaScript component records timing, clicks, pointer movement, scrolling, touch, keyboard behavior, and basic browser signals—not what a respondent writes.
- 02
Analyze afterward.
Once the response is complete, Ramon’s proprietary model processes the full behavioral sequence. The survey experience remains untouched.
- 03
Receive a Human Score.
The completed response receives one understandable signal that research teams can use in their existing quality workflow.
Inside the model
Purpose-built evidence, not a generic AI wrapper.
Ramon’s model was trained specifically on behavioral sequences from online survey sessions. It looks for temporal patterns associated with known AI-driven interactions and combines the evidence into one readable Human Score.
256-event windows
Long sessions are evaluated as ordered windows, preserving both the sequence of events and the time between them.
Known-AI patterns
Training includes survey sessions known to have been completed by AI-driven agents alongside unlabeled survey sessions.
Conservative by design
The result summarizes behavioral compatibility. It does not claim to identify a person or deliver a binary verdict.
Answer text and personal form entries are not model features. Ramon evaluates the interaction around the response, not the response itself.
Post-survey
Post-survey by design.
Ramon is deliberately not an ahead-of-time filter, a just-in-time intervention, or a real-time gate. It adds evidence when the full interaction is available.
No interruption. No live verdict. No influence on the respondent.
Book a demo