The hardest part of building Hoovik — my open-source AI-powered meeting platform — wasn’t WebRTC signaling or media pipelines. It was managing real-time multimo
Anupam Kumar, the creator of the open-source AI meeting platform Hoovik, found that the most challenging aspect of development was not the core WebRTC technology but managing real-time multimodal AI inference. This involved complex coordination of PyTorch, MediaPipe, and AudioWorklets across distributed services. Kumar aimed to achieve this without compromising performance through event loop blocking or memory exhaustion, especially when dealing with unstable network conditions and disappearing media streams. AI
IMPACT Highlights the complex infrastructure challenges in deploying real-time multimodal AI for applications like meeting platforms.
