We're building the prediction layer for advertising.

Origins

Cortex's model began as a clean-room implementation of public neuroscience research, trained on independent open data. We never train on customer ads. We only use them to compute a per-brand signature at inference time.

That separation is the whole point. A tiny DTC brand and a multi-billion-dollar holding company can both use Cortex without either of them ever seeing the other's creative.

Principles

• Predict the thing, don't predict a proxy of the thing.
• Show your math. The customer should be able to audit any score.
• Free failures. If a scoring doesn't run, it doesn't get billed.
• One screen. Not eight tabs.
• The model is our moat, but the data is the customer's.