Harbor - Bridging language barriers in America
Team led by Extralit founder Jonny Tran, Ph.D. (UT Arlington): AI/LLM engineer experienced with PyTorch, Transformers, graph ML, bioinformatics, FastAPI, scalable Python pipelines.
Project Description
Harbor is an AI-powered translation companion specifically designed to empower non-English speaking immigrants in the United States by transforming complex communications into clear, actionable guidance in their native language. Unlike traditional translation tools, Harbor operates through a simple email forwarding system where users can send important documents like bills, collection letters, or government notices to language-specific addresses ([email protected], [email protected]) and receive comprehensive responses that include not only accurate translations but also cultural context, deadline identification, and actionable next steps. Built with accessibility at its core, Harbor eliminates barriers by requiring no app downloads, account creation, or technical knowledge. Users simply forward emails and receive intelligent assistance that helps them navigate American bureaucracy with confidence. The service uses Mailparser, Postmask and OpenAI GPT-4o to deliver contextually aware translations that address the high cognitive load immigrants face when dealing with unfamiliar legal and administrative systems.
Backend Email Processing System
- Node.js/Express server with Postmark webhook integration for real-time email reception
- Mailparser library extracts email content, subjects, and sender information from incoming messages
- Language detection automatically identifies target language from recipient address prefix ([email protected] → Spanish)
AI Translation & Analysis Engine
- OpenAI GPT-4 integration with specialized prompts: “Translate and explain this {document_type} in simple {target_language}. Highlight deadlines, required actions, or important warnings. Include cultural context where relevant.”
Response Generation & Delivery
- HTML email templates create mobile-responsive, beautifully formatted responses with clear sections
- Structured output organizes information into: original text, translation, summary, action items, and cultural context
[only the frontend work, the backend currently doesn’t work]