Why we built it in-platform
Every other dating app hands you a match and tells you to go figure it out. You find a time, you share a Zoom link, you hope for the best. The app learns nothing from what happens next.
Attune's video date environment exists for two reasons. The first is practical: an in-platform date is a better experience — no setup, no link-sharing, no external account required. The second is architectural: it's the only way to capture interaction data that actually teaches the system what a good match looks like in practice.
You can't get emotional synchrony data from a Zoom recording you don't own. You can't capture turn-taking balance from a platform you have no access to. The in-platform date is not a convenience feature. It is how Layer Three of Attune's system works.
Consent, before anything else
This is not a buried checkbox in terms and conditions. Biometric and behavioural data requires an active, informed decision from every person involved. The consent screen is a mandatory step in the date flow, not an optional extra.
How the video date works
Both people agree on a time through the Attune messaging interface. A video date is proposed; both accept. No external scheduling tool. No email exchange.
Before the call opens, each person individually reviews and accepts (or declines) the analysis consent. The date can only begin when both have responded.
A private, two-person video call inside the app. No other participants. No recording indicator visible to either party during the call — you're having a conversation, not performing for a camera.
Each person privately answers: second date, yes or no? Neither answer is revealed to the other person unless both say yes. If only one says yes, neither is told which way. The outcome is sealed.
The recording is processed after the session. Features are extracted. Raw video is deleted within 48 hours. The interaction features and the outcome label are stored together, anonymised, as a training data point.
What the analysis captures
Post-session analysis extracts interaction features — not content. The system does not transcribe or analyse what you said. It analyses how the conversation happened.
AU patterns
Smiles, tension markers, negative affect indicators — tracked across the full arc of the conversation. The trajectory matters, not just the peak.
Emotional synchrony
How closely expressions align between participants over time. Interpersonal synchrony is a strong predictor of felt connection — documented across multiple research contexts.
Conversation balance
Turn-taking patterns and listen-to-speak ratio. Persistent imbalance in conversational dynamics reliably predicts lower satisfaction with the interaction.
Emotional trajectory
Did valence and arousal increase or decrease as the conversation progressed? An improving arc is a meaningfully different signal to a declining one.
These features are stored alongside the mutual outcome label — yes/yes, yes/no, or no/no. That labelled pair is what the Attune learning system trains on. Over time, the model learns which patterns of interaction tend to produce mutual interest, and weights future matching accordingly.
The double-blind mechanism
The post-date feedback step is the core of what makes Attune's dataset useful. The mechanism is deliberately simple: one question, one binary answer, sealed until mutual.
The double-blind design is not just a nice user experience. It is a data quality decision. Social desirability bias — the tendency to say what you think is expected — massively distorts self-reported dating outcomes when responses are not confidential. A person who knows their match said yes is far more likely to reciprocate regardless of how they actually feel. A person who knows their match said no is far more likely to suppress a genuine yes.
By sealing responses until mutual, Attune eliminates both distortions. What the system receives is the honest answer — not the socially managed one. That makes the outcome label reliable as training data.
Data handling
- Raw video from each session is retained for a maximum of 48 hours post-session, then permanently deleted
- Extracted interaction features are anonymised before storage — they cannot be linked back to the recording
- Mutual outcome labels are stored alongside features, but neither participant's response is individually identifiable in the training data
- Users can request deletion of all interaction data from their account at any time — this includes all features extracted from sessions they participated in
- Attune is fully compliant with UK GDPR including the right to erasure and data minimisation requirements for biometric data processing
Dates that actually tell us something.
Video dates are how Attune gets smarter. The earlier you join, the more you shape what the system learns. Launching UK Q3 2026.
Join the waitlist How the full system works →