LangBridge AI

Real-Time Multilingual Marketing Insight & Content Agent

Comment

Inspiration

In today’s global market, brands must speak the language, literally and culturally, of diverse audiences. But it’s hard for marketers to know what Gen Z in Japan or professionals in France are saying about products, trends, or campaigns. We wanted to build an agent that not only understands real-time international sentiment but also generates localized, high-quality marketing content all automatically.

What it does

LangBridge AI is a real-time multilingual marketing agent that:

  1. Takes user input like a product/topic and target regions
  2. Scrapes real, live web content (e.g., reviews, Reddit, marketplace data) using Bright Data MCP
  3. Uses Amazon Bedrock LLM (Claude) to craft smart, culturally adapted prompt instructions
  4. Sends these prompts to the MiniMax API, which generates localized ad copy and creative media briefs

How we built it

Frontend: A lightweight React form where users select product, countries, tone, and platform (e.g. “Instagram ad for Gen Z in France and Japan”)

Backend: Python-based FastAPI server to orchestrate the pipeline

Bright Data MCP: Used their MCP Server API to pull live search results and snippets for product-related queries in each target country

Amazon Bedrock: Leveraged Claude or Command R+ to parse the scraped content and generate structured instructions (e.g. “Make a 15s Gen Z-friendly TikTok ad with humor in Japanese”)

MiniMax API: Used those structured prompts to generate the final ad copy in each language, styled appropriately for each platform and audience

Challenges we ran into

Multilingual search scraping: It was challenging to craft reliable queries for different languages and extract meaningful snippets under rate-limiting conditions

Prompt design across two LLMs: Claude and MiniMax have different input expectations, so we had to fine-tune the way prompts were passed from one to the other

Maintaining tone + cultural nuance: Getting the ad copy to sound natural and locally relevant took trial and error in prompt tuning

Accomplishments that we're proud of

We built a fully functioning end-to-end pipeline that scrapes real data and returns localized ad content in under a minute Created sample ads tailored to distinct tones.

What we learned

LLMs like Claude are great for abstract reasoning, while tools like MiniMax excel at media-specific generation Building a truly multilingual agent requires more than just translation — it needs regional context, voice, and purpose

What's next for LangBridge AI

Add voice interaction with voice AIs for conversational interactions Implement ad performance feedback loop, where successful campaigns can retrain prompt styles Can make voice-lip synchronized output

Built With

Share this project:

Updates