Translate tech talks using open-source models and cloud-native Wasm

Vivian Hu - Second State

Language barriers often limit the global reach of valuable technical content. LLM agents seems to be a promising technologies to solve the problem.

Let’s think about a seamless pipeline powered by open source models: Whisper for technical speech recognition, Qwen-72B for context-aware translation, and GPT-SoVITS for natural speech synthesis. To run these models easily and scalable, we use CNCF WasmEdge runtime serves as a unified runtime, enabling efficient deployment and operation of these AI models. Moreover, we can use one single Wasm API server to run multimodal models, which is openAI API compatible.

Attendees will learn practical patterns for deploying multiple AI models with WasmEdge and strategies for maintaining technical accuracy in AI-powered translation. We’ll dive into implementation details including model optimization and prompt engineering, and workflow integration patterns.