Building Real AI Products: Balancing Vibe, Security, Quality, and Speed
Agenda:
5:00 pm to 5:15 pm - Introduction and Networking
5:15 pm to 6:00 pm - Panel Discussion on Building Real AI Products: Balancing Vibe, Security, Quality and Speed
6:00 pm to 6:20 pm - From research to production: a case study of AI translations at Meta. Tech Talk.
6:20 pm to 6:30 pm - Networking
Join us for a panel discussion where our executive panelists will discuss the choices and decisions that they have had to make in building real AI products at scale. Panelists include:
Sandeep Kaipu (Broadcom)
Mayada Abdelrahman (Cisco)
Sarah Feldman (ex-Google, Tesla)
Sean Roberts (ex-Yahoo)
Tech Talk: From research to production: a case study of AI translations at Meta
The talk dives deep into the development and implementation of a large-scale media processing pipeline at Meta, designed to enable automatic dubbing and lip syncing for media content. The pipeline leverages advanced AI to perform real-time translation and voice synthesis, allowing for seamless multilingual content delivery across Meta’s platforms.
The talk addresses the technical challenges encountered in scaling such a system. These include managing inference latency to ensure quick processing, efficiently scheduling tasks across distributed infrastructure, and achieving accurate voice isolation to separate speech from background audio. The team also delves into the complexities of media timing and alignment, which are critical for synchronizing translated audio tracks with the original video’s lip movements and timing.
The system supports alternate audio track delivery, enabling users to select different languages or voices as needed. Instrumentation and monitoring are emphasized as essential for maintaining system reliability and performance at Meta’s scale. In the talk, the speaker will present Meta's approach to model evaluation, ensuring that the AI-generated translations and dubbing meet high standards for quality and naturalness.
Overall, the talk aims to provide a comprehensive overview of Meta's engineering solutions and innovations to enable high-quality, scalable AI-powered media translation, highlighting the technical hurdles and strategies used to overcome them.
Who should attend:
Engineers, Developers
Architects
Engineering Managers
Director of Engineering
VP of Engineering
CTO
Highlights from previous event