Since 2026 · product · ml · speech · logistics
VoxTruck
An English learning platform built for truck drivers, taught in the vocabulary of the job rather than the classroom.
Generic language apps teach you to order coffee. Drivers need to handle a dispatcher, a border officer, and a DOT inspection.
What it is. A language platform that teaches drivers English inside the situations they actually hit (dispatch calls, border crossings, paperwork, inspections) with speech scoring and lesson difficulty that adapts to the individual driver.
Why it matters. For a cross-border carrier, a driver's English is an operational variable, not a personal enrichment goal. It shows up in delays, misread instructions, and inspections that go longer than they should.
Payoff. In internal pilot with TBM Carriers drivers. Too early for outcome numbers, and I'd rather say that than invent one.
What this actually is
VoxTruck is an English learning platform for professional drivers, built with TBM Carriers.
The premise is that context is most of the value. A driver doesn't need a general A1-to-C1 ladder; they need to be fluent in the forty situations that make up their week, and confident enough to speak in them under pressure.
So the content is logistics-native: the phrasing a dispatcher actually uses, the questions an officer actually asks, the documents that actually get handed across.
Where the machine learning does work
Both exist for the same reason: a driver's schedule doesn't accommodate a fixed class. The learning has to fit into the gaps, and it has to be useful without a teacher present.
- Speech scoring: the driver speaks, and the model scores pronunciation and fluency, so practice doesn't require a human on the other end
- Adaptive difficulty: the system tracks what an individual driver keeps missing and reshapes what it serves next, instead of marching everyone through the same sequence
Honest status
Internal pilot with the company's own drivers. Contained on purpose. A language product with confident-sounding wrong feedback is worse than no product, so the scoring gets tested against real drivers before it gets tested against a market.
In pilot. Numbers when there are numbers worth reporting.