Inspiration

Most of us have been there in another country. You know a few phrases, you walk up to the counter, someone asks you something quickly, and you freeze. You end up pointing, saying one word, or walking away. Language apps teach you vocabulary, but the hard part isn’t the words. It’s the moment itself. One “Sorry?” from a stranger is enough to make you panic. We wanted to help people have a real conversation with a real person: face to face, with eye contact, in their own voice.

What it does

Tenin is an AR companion that helps travelers rehearse a conversation right before it happens. Most VR language apps put you in a virtual café with a virtual waiter. You get good at talking to avatars, but the real world still feels just as scary. Tenin is built around the opposite idea: the goal isn’t to practice, it’s to actually speak. It starts from where you are. Tenin animates a photo of the exact place you’re about to walk into, minutes before you walk in. You see what’s coming before you speak. We show the likely outcome first, then your line. Knowing what the other person will probably say takes away most of the fear. AR keeps you in the world. You can still see the room, make eye contact, and use your own voice. Tenin adds just enough to help, then gets out of the way.

How we built it

Hardware: XREAL Air 2 Ultra glasses connected by cable to an Android phone running our Unity app. The phone handles everything between the glasses and the cloud. OpenAI: plans lessons, suggests situations from the photo, judges whether the user was understood, and writes the reflection and glossary. Cartesia: voices the other person and plays phrase audio. Visko: turns a chosen scenario, or the user’s own photo, into a scene video. Meshy: generates the 3D rooms for Pre-existing Lessons, exported into Unity. Privacy on the phone: faces are blurred and location data stripped before any photo leaves the device. If the face detector fails, nothing is sent.

Challenges we ran into

Limited hardware. The only AR glasses available to us were the XREAL. Rather than change our idea, we bought an Android phone to develop and test on. Slow video generation. Each prompt takes about 10 seconds to generate. We made the wait part of the lesson: a calm breathing step and a preview of your next line instead of a loading bar. Limited display space. We kept line cards small and low in your view so they never block the real person in front of you.

Accomplishments that we're proud of

The full Live Lesson flow working end to end: photo → suggestions → animated scene → rehearsal → speak for real. Designing for anxious beginners. The user starts every lesson and chooses every move, and there’s always a one-tap exit. Privacy built in from the start: no faces or location data leave the phone, and images are deleted after each lesson.

What we learned

The fear is less about vocabulary and more about not knowing what comes next. Showing the likely outcome before the line made rehearsal feel safer. When AI handles a high-stakes topic, it has to say “ask,” never “safe.” Allergy and health lines always prompt the user to check with staff or a clinician.

What's next for Tenin

Lighter glasses. We built on the XREAL because it was the hardware available here. The real target is lightweight everyday glasses like the RayNeo X3 Pro, which have their own camera and are light enough to wear around a city. That means designing for a smaller field of view and shorter battery life. Finishing the glasses screens for the spilled-drink, supermarket, and doctor scenarios. User testing to measure anxiety before speaking, how often people actually speak for real, and how often they’re understood on the first try. More languages after Japanese.

Built With

Share this project:

Updates

Submission history