Building the world’s first EQ models.
We train large models to understand and persuade people. The signal: over a million live conversations, each labelled by what the human did next.
BACKED BY
Seedcamp
QuietCarles Reina
We train large models to understand and persuade people. The signal: over a million live conversations, each labelled by what the human did next.
SeedcampQuietFrontier models are trained on the public internet. They can describe a product perfectly. But they have never sat with ten thousand hesitant buyers, so they cannot tell which reply moves a person forward and which one ends the conversation.
Most of what closes a sale never appears in the text at all: whether to mention price, when to follow up, when to go quiet. It is not written down anywhere a model could read it.
We have that data, because our systems run these conversations end to end, and every thread comes back labelled with what the customer actually did.
It learns from the one signal the public internet does not contain: what a real buyer did next. Every thread our systems run comes back labelled with the outcome, and that label is the training signal.
We judge it the way the merchants who pay us do. Not on a benchmark, but on what the conversations earned.
A live A/B test on real customers, across more than 3 million conversations. The only variable: which model wrote the replies.
50,000+ live conversations a day, and no management layers. Everyone reports to the founder.
