What you need to write about Frank.
A boilerplate in three lengths, the facts with their source declared, a link to the team by name and role, and an email a person answers. Every fact here has a source and a date.
Frank, in three lengths.
Paste whichever fits. All three say the same thing at different levels of detail, and none needs our correction.
- Frank trains difficult work conversations by voice, with an AI counterpart that answers back, pushes, and can end the meeting.
- Frank is a voice-based human skills training platform. Teams practice the hard conversation before having it, against an AI counterpart that responds in real time, and each person gets a private report at the end on how they held it.
- Frank is a voice-based human skills training platform built in Concepción, Chile. It puts an AI counterpart on the other side of the conversation: one that answers back, pushes, interrupts and can end the meeting, with the same friction that later shows up in the room. When the session ends, each person receives a private report with paralinguistic analysis: not what they said, but how they said it, compared against their own baseline. A session lasts 5 to 10 minutes, runs in the browser and is available in Spanish and English. In 2026 the team won Red Bull Basement Chile and placed in the world Top 40 out of 140,000 entries.
The name is written Frank, with a capital F and never in all caps. The rules for the logo, the palette and the typography are in the brand kit.
The facts, with dates.
What the product does today and what happened in 2026. Photos and how to check each competition are on the Backed by page.
Voice sessions of 5 to 10 minutes, in the browser.
National champions of Red Bull Basement Chile.
Top 40 in the world at Red Bull Basement, out of 140,000 entries.
Finalists at the 7th Gear Challenge.
A team of four in Concepción, Chile.
What you can cite, and where it comes from.
Three things a story usually needs, already settled here, each one with a way to check it.
- The awards, with an outside jury, a date and a place. Red Bull published the national final result.
- The report each person receives: their delta against their own baseline, with signal quality stated in it. You can see one in a demo, and it is never a ranking between people.
- Engine fidelity against openSMILE eGeMAPS v02 and Praat on the RAVDESS corpus, 24 speakers: .99 on the pitch family, .97 on loudness, .94 on harmonic quality. This is an internal fidelity measurement of the engine, not validation of outcomes: it does not imply these signals predict performance.
All three come with their source: a public source, a stated method and the product itself running.
The team.
Four people. The names, the roles, the portraits and the context are on the team page.
Brand downloads.
The logo in its three versions, the palette with hex values, the typography and the usage rules are in the brand kit. The logo files are here too, in case that’s all you need.
