Lingo

• 

  Our project in use.

Inspiration

We’ve all been there: a joke lands flat, a "simple" instruction causes chaos, or a slang term leaves us feeling isolated. This project was inspired by a real-life texting miscommunication where a simple idiom can cause total confusion. We realized that while translation tools handle words perfectly, they often fail at meaning. We wanted to build something that translates culture, not just language.

What it does

Lingo is an AI-powered intelligent "Cultural Translator" for real-time nuanced messaging.

It acts as an invisible interpreter that sits between a sender and a receiver to prevent miscommunication caused by idioms, slang, and corporate jargon.

For the sender:

It analyzes messages before they are sent. If a user types something culturally specific (e.g., "Let's not boil the ocean" or "No cap"), the app:

• Flags it: Warns the user that the phrase might be confusing to a non-native speaker.
• Suggests a Fix: Uses AI to propose a culturally neutral alternative (e.g., changes "boil the ocean" to "waste effort").

For the receiver:

If the sender chooses to keep the original idiom, Culture Bridge assists the receiver by:

• Highlighting the Phrase: It identifies the specific confusing words in the chat bubble.
• Providing Context: The receiver can tap the highlighted phrase to see a "definition card" explaining what it actually means in that specific cultural context.

How we built it

Frontend: Built with React.js (Vite) and Node.js. We used Tailwind CSS for styling and Framer Motion to create the smooth "iOS-style" animations (like the Dynamic Island and bottom sheets).

AI & Logic: The core intelligence is powered by Google Gemini (via OpenRouter). We engineered a strict system prompt that forces the LLM to act as a "Sociolinguistic Expert," returning data in a precise JSON schema containing phrase, definition, context, and suggestion.

Backend: We utilized FastAPI as our middleware layer. It handles the request traffic, manages the API keys, and parses the JSON response from the LLM before streaming it back to the client.

Dev Tools: We accelerated our UI development using v0 by Vercel to prototype components rapidly.

Challenges we ran into

The "Latency vs. Accuracy" Trade-off: Calling an LLM for every keystroke is too slow. We had to implement a "debounce" strategy where the AI only analyzes the text when the hits send.

Parsing Multiple Idioms: Initially, our code crashed if a user used two idioms in one sentence (e.g., "Break a leg but don't cut corners"). We had to rewrite our backend to return an array of matches and create a dynamic Regex engine on the frontend to highlight multiple distinct phrases simultaneously.

Contextual False Positives: Teaching the AI that "Break a leg" is an idiom in a theater context, but a medical emergency in a hospital context.

Accomplishments that we're proud of

• We eliminate the miscommunication gap between different cultures, age groups, and languages.
• We bridge the gap between fluent English (textbook grammar) and cultural English (how people actually speak), making it useful for:
  • Global Remote Teams: Decoding business jargon.
  • Immigrants & Expats: Understanding local slang.
  • Healthcare: Ensuring clear instructions between doctors and patients.
• We successfully simulated a dual-phone environment in a single browser window to demo the "Sender" and "Receiver" experience simultaneously.
• We successfully engineered a prompt that stops the LLM from hallucinating and forces it to return clean, parsable code every time.

What we learned

• The difference between a buggy app and a great one is often how specifically you instruct the model's "System Role."
• We gained a deeper appreciation for how difficult English is to learn, and how powerful LLMs are at bridging that gap.

What's next for Lingo

-EI & EQ friendly interface

• Browser Extension: Moving beyond the demo app to a Chrome Extension that overlays on top of Gmail, Slack, and WhatsApp Web.
• Tone Detection: Adding "Sentiment Analysis" to warn users if their message sounds accidental aggressive or passive-aggressive in a different culture.
• Enterprise Integration: A Slack Bot version for international companies to install in their workspaces. Especially B2B and B2C

Built With

• fastapi
• gemini
• react