Senior Machine Learning Engineer (Voice & Video AI)

Join Telnyx as a Senior Machine Learning Engineer (Voice & Video AI) and be at the forefront of building the next generation of conversational voice and video AI. In this hands-on applied AI role, you will have end-to-end product ownership, taking promising models from evaluation and fine-tuning into production. You will work with state-of-the-art GPU hardware and contribute across the stack, from inference services to backend APIs and frontend customer portal changes.

- Concevoir et développer des modèles de voix et de vidéo de pointe, y compris des agents conversationnels capables de voir, d'écouter, de parler et de répondre naturellement en temps réel. - Assurer la mise en production de modèles prometteurs, en optimisant les performances de diffusion, l'utilisation du GPU, la latence et le coût de service tout en maintenant la qualité de la conversation. - Posséder des fonctionnalités depuis les expériences de modèle jusqu'aux services backend, aux API, aux modèles de données et aux interfaces du portail client.

- End-to-end engineering ownership: You’re comfortable working beyond the ML codebase—making backend and frontend changes, navigating unfamiliar repositories, and following a feature through to a working customer experience - An agent-driven development mindset: You actively want to work with the latest coding agents and harnesses, improve your workflows as the tools evolve, and apply strong engineering fundamentals to verify and take responsibility for their output - Strong systems judgment: Experience with streaming or latency-sensitive systems, rigorous evaluation, and the tradeoffs between model quality, responsiveness, reliability, and cost. You can turn research into practical engineering decisions - Deep ML expertise: 6+ years in machine learning or related engineering, with production experience and depth in speech generation, speech-to-speech systems, video generation, or multimodal models. Strong Python and PyTorch skills and hands-on experience adapting pretrained models