ParaFrame – Parallel Agent Framework for Long-Video QA
A parallel agent-based framework for efficient long-video question answering built on a Mixture-of-Experts (MoE) architecture.
ParaFrame is a parallel agent-based framework for efficient long-video question answering.
It leverages a Mixture-of-Experts (MoE) architecture to distribute reasoning workloads across specialized agents, enabling scalable processing of long, complex video content while maintaining answer quality.
The project showcases system-level design for multimodal reasoning, including agent orchestration, parallel execution, and efficient use of video and language models.