This paper presents a comprehensive approach for analyzing financial news in Portuguese and extracting sentiment metrics using the LLM Gemini Pro model and data from the Ibovespa index. The study describes the process of creating guidelines customized to annotate news sentiment and recognize key entities. Three different methods for generating sentiment metrics are introduced: a direct metric approach, a seven-day moving average, and an exponential moving average. Additionally, the paper elaborates on developing filters designed to reduce noise and enhance indicators through entity-based and topic-based filtering techniques. Statistical analysis, including correlation coefficient calculations across various scenarios, reveals the efficacy of entity-based filtering in establishing a strong correlation between sentiment indicators and daily market reactions. Furthermore, significant correlations between moving average and exponential moving average indicators and future market trends underscore their predictive capability.