Inspiration
Many grandparents who don't speak English immigrated to the US. Most of the time, their children struggle to communicate them, using google translate on messages to understand what they are saying. Living in the US is also hard, as the grandparents need to navigate life in a new language they don't fully understand.
What it does
Bridge is an iMessage agent for families who don’t share a language. Each person texts Bridge in their own language, and Bridge relays the message to the rest of the family, translated with the right tone: respectful to elders, casual between cousins. Replies, tapbacks, edits, unsends, photos and voice notes all carry over. When a phrase is ambiguous, like an idiom, sarcasm or a tone-dependent “whatever”, Bridge doesn’t guess. It privately asks a bilingual relative to pick the right meaning with a tapback, and learns from the answer. Emergencies like “help, I fell” skip review and reach everyone immediately.
Bridge also shops for the people you love. When Grandma mentions a need, Bridge finds options that fit her, shows the whole family the same cards in their own languages, lets everyone vote with a tapback, checks out with Visa, and sends order updates. No new app: it all happens in the family’s existing chats.
How we built it
We built Bridge in TypeScript/Nodeon Photon 2.0’s Spectrum SDK. Each family member has a one-on-one iMessage thread with Bridge, which relays every message into everyone else’s thread. A message map keeps replies, tapbacks, edits, and unsends in sync across all of them. Gemini translates with relationship-aware tone and a confidence score. When the score is low, Bridge asks a bilingual relative to confirm with a tapback and learns from the answer. Voice notes are converted with ffmpeg and transcribed by Gemini. Photos are explained in each reader’s language. For shopping, SerpApi Google Shopping finds products, the AI ranks them for the person, and the family votes with tapbacks. Checkout uses Visa Direct and the Foreign Exchange Rates API, both in the sandbox. Two of us built it in parallel, with clear file ownership, a terminal simulator, and offline tests, using Claude Code throughout.
Challenges we ran into
Moving to Photon 2.0 mid-hackathon. We first built on Photon’s older iMessage kit, then switched to the Spectrum SDK to meet the track’s requirements. Spectrum doesn’t deliver incoming edits, unsends, or tapback removals, so we read those from the underlying event stream. We also learned that Photon’s shared pool gives each user their own Bridge number, which changed how we began testing.
Accomplishments that we're proud of
It works on real phones, not just in a demo video. Two iPhones texting in Chinese and English, with every reply, tapback, edit, unsend, photo, and voice note mirrored correctly across threads, on Photon 2.0.
What we learned
Translation is about meaning, not words. The same phrase can be loving or annoyed depending on tone. We learned to treat confidence as a first-class output and to bring in a human exactly when the meaning is uncertain.
What's next for Bridge
Next, Bridge will go beyond translating words to bridging cultures. It will explain the meaning behind phrases (“Have you eaten?” means “I care about you”), coach etiquette and honorifics, mark holidays like Lunar New Year, and respect dialects like Cantonese. We also plan to add WhatsApp, voice calls, and Visa agent payments.
Log in or sign up for Devpost to join the conversation.