Hi, I'm Mark Duppenthaler

Omnilingual ASR Visualization

As the lead developer, built an immersive 3D language visualization demo for Meta's new ASR (automatic speech recognition) model, showcasing 1600+ supported languages through an interactive WebGL-powered globe visualization. Users can explore languages geographically and see the unprecedented global coverage of the model. Visit the demo to explore all supported languages across the world. This model especially helps underrepresented languages gain accessibility to speech recognition technology for the first time.

Omnilingual ASR Transcription Tool

As the lead developer, built a hosted Docker space on HuggingFace that transcribes video and audio files with an interactive editing interface for refining AI-generated transcriptions before export. Leverages Meta's Omnilingual ASR model supporting 1600+ languages to provide accurate transcription across a vast array of languages, especially underrepresented ones. Visit the demo to run transcriptions on your own media, though you'll need to clone the space for the full unrestricted experience.

Bouquet

As the sole developer, built a full-scale translation crowdsourcing platform to improve machine translation models across 4000+ underrepresented languages. Designed and implemented the complete stack including user authentication, an intuitive translation collection interface, and a robust Flask backend with PostgreSQL database. Architected on AWS using CloudFront, WAF, API Gateway, ALB, ECS, ECR, Cognito, SES, Postgres RDS, Secrets Manager, CloudWatch, and Google Analytics for a secure, scalable, and production-ready system. Visit Bouquet and contribute translations to help improve machine translation in your language!

Seamless Streaming

As the sole developer, built an immersive Oculus experience using a streaming version of the expressive translation model, enabling real-time bidirectional conversations across language barriers with low-latency audio processing. Users could engage in natural conversations with people speaking different languages while seeing live transcriptions overlaid in their headset view. Demoed this groundbreaking technology to international press across multiple countries, including a showcase in Japan that aired on Japanese national television. Try the HuggingFace demo (may need to clone the space) to experience the streaming translation technology yourself.

Seamless Expressive

Built the interactive demo for breakthrough research in expressive speech translation that preserves vocal style, emotion, and prosody across languages—delivering translations that sound natural and emotionally authentic. This research became the foundation for Instagram's voice translation feature, bringing expressive multilingual communication to millions of users worldwide. Published in a Meta AI paper, this work demonstrates how AI can maintain human expressiveness in translation.

Seamless M4T

Implemented critical features for the interactive demo showcasing a groundbreaking multimodal translation system that supports speech-to-speech, speech-to-text, text-to-speech, and text-to-text translation across nearly 100 input languages to 35 output languages. Built the user interface and integrated ML models deployed on AWS SageMaker. Published as a research paper and released as an open-source model, enabling seamless communication across language barriers through advanced AI-powered translation technology.

Ana Carolina ♥ Mark

When your fiancée asks if you can build a wedding website, you say yes. When she's Brazilian and the wedding is in Brazil, you make it bilingual. When you're a developer, you probably spend way too much time perfecting the animations. Built with modern React architecture featuring full internationalization (i18n), responsive design optimized for mobile guests, and smooth page transitions. Includes RSVPs, event details, travel info, photo gallery, and a beautiful prewedding video. The most important product launch of my life.

Facebook API Upgrade Tool

Built the official tool from Facebook for helping developers upgrade their apps to new versions of the Graph API. Personalized for each developer from real calls coming from their app, showing which calls would break with upcoming API changes.

Built the data pipeline which aggregated trillions of API calls daily, analyzed them for potential breakages, and generated personalized reports for each developer. This project improved developer experience and reduced support costs significantly. Patent awarded.

Facebook Anonymous Login

As the sole engineer, built this product announced at F8 2014 by Mark Zuckerberg, which made headlines worldwide for allowing users to log into apps anonymously. Designed and implemented sophisticated infrastructure on top of Facebook Login that maintained user anonymity within an inherently social platform—a significant technical challenge that required rethinking authentication and data access patterns. My first patent was awarded for this innovation.