Mundo AI
Prepare data and test how models handle audio and video.
Mundo AI provides datasets and evaluation tools for AI labs working on audio, video and other non-text inputs. Its work focuses on how models interpret those inputs.
Mundo AI
Founded 2024 · Vancouver, British Columbia, Canada
Official WebsiteUS$20 million
Mundo raised the round as it built datasets and evaluations for AI systems that work with speech, video and gestures.
Led by GreatPoint Ventures. Participation included Y Combinator.
Source · thesaasnews.com ↗Mundo’s founding team divided responsibility across company leadership, engineering, product, and operations. The company builds training data and evaluations for models that process audio, video, and other inputs. The founding team included Naijide Anwaer (COO), Garreth Lee (CPO), and Kenneth Wu (CTO).
Prepare data and test how models handle audio and video.
Mundo contributed audio to the Smart Turn training dataset and later worked with Sesame on TurnBench. Those projects show two sides of the same problem: providing examples that teach a voice model when people finish speaking, and testing whether the model actually learns when to take its turn.
Mundo creates data and evaluations for systems that need to interpret audio and video. Mecka collects human movement and behavior specifically for robotics. The overlapping question is how to gather examples of real-world behavior, but the model outputs to be evaluated differ.