Omar’s Learning Spanish with Google’s Flow & Omni Flash
Client: Self-initiated
Project: Learning Spanish with Google Flow and Omni Flash
Role: Director + Student
Production: Brainclub 5000
Deliverables: 90 personalised short films
Timeline: Five weeks
Five weeks in Las Palmas, zero Spanish. So I turned the problem into a brief: could social content itself become the teaching surface?
I used Google's Flow Agent to research me first, my work, my humour, my life, then had it write, cast and generate a series of short form clips teaching Spanish phrases with their English translations. Thirty personalised films, and the phrases get stranger as they go, on purpose: new and novel information is easier to remember, so the absurdity is the pedagogy. A phrase you laugh at is a phrase you keep.
The result is a language course that behaves like a feed. Watchable in the moments you would be scrolling anyway, funny enough to rewatch, and built for one specific student, which is the part that matters. This is what generative systems change about learning content: the research, the scripts, the casting and the films all came from one agent working from one person's context. A first, as far as I can find, in using an agentic pipeline and Omni Flash as a learning surface.
And it worked. I ordered the coffee.
I used Google's Flow Agent to research me first, my work, my humour, my life, then had it write, cast and generate a series of short form clips teaching Spanish phrases with their English translations. Thirty personalised films, and the phrases get stranger as they go, on purpose: new and novel information is easier to remember, so the absurdity is the pedagogy. A phrase you laugh at is a phrase you keep.
The result is a language course that behaves like a feed. Watchable in the moments you would be scrolling anyway, funny enough to rewatch, and built for one specific student, which is the part that matters. This is what generative systems change about learning content: the research, the scripts, the casting and the films all came from one agent working from one person's context. A first, as far as I can find, in using an agentic pipeline and Omni Flash as a learning surface.
And it worked. I ordered the coffee.