In the cryptocurrency field, sentiment analysis systems face challenges due to inconsistent interpretation of multilingual news. To achieve quick and accurate market insights, the team adopted a "multi-model consensus" architecture, designing a dual-path asynchronous pipeline to ensure efficient processing. The new system utilizes domain-specific models and introduces a lightweight arbitration layer to optimize data processing and evaluation, reducing operational costs and enhancing stability. The future goal is to evolve into an intelligent agent capable of tracking asset tokenization narratives, driving new financial infrastructure.