How KLIPY Ended Up on the Google I/O Main Stage

Last month, three startup logos lit up the main stage at Google I/O 2026 in Mountain View. One of them was ours.

During the developer keynote, Logan Kilpatrick - who leads much of Google’s developer-facing work on its Gemini models - pointed to KLIPY alongside Ramp and Resemble AI as examples of companies building on Google’s newest AI models. Here’s how we got there.

A quick reminder of what KLIPY does

KLIPY is the Expression API layer for short-form visual content - that powers more than 8000 apps including BeReal, Microsoft Swiftkey and Baidu. We give platforms a culturally relevant, localized content library they can drop in from day one, and a monetization engine that turns a feature most apps treat as a cost center into a revenue stream. Today that infrastructure serves billions of content requests every month.

That scale is exactly what made working closely with Google’s teams a natural fit.

Building with Google’s newest models

KLIPY bet early and heavily on Google’s AI labs and their latest models, and we put them to work across the product. Instead of treating this as a side experiment, we used it to make the whole company more modern, more innovative, and more efficient - from how content gets understood and categorized, to how it gets matched to the right user in the right context, to how we keep our infrastructure lean.

Keeping content safe for everyone

The part we’re most proud of, though, is harder to fit on a slide: how we keep our content compliant and properly rated for every audience that touches it.

On KLIPY, short-form media is published to thousands of apps representing scores of user groups in the blink of an eye. That makes content safety non-negotiable. Our approach pairs automation with human moderators, and the order matters. Our automated systems filter out NSFW material, flag edge cases, and rate content so the right things reach the right user groups. But the last call is always made by a human moderator at KLIPY.

Automation assists our moderators; it never replaces them. Every gray-area decision goes through a person, because automated systems are powerful but not infallible, and the cost of getting safety wrong is too high to delegate entirely to a model. That human-in-the-loop layer is what lets our partners trust KLIPY content inside their own apps.