This paper presents a fully automated pipeline that collects, processes, and models public Instagram data to quantify hate directed at high-profile accounts, as well as hate disseminated by them. A headless Selenium crawler gathers the fifteen most recent posts per influencer, including images, captions, up to 2,000 comments, and basic metadata, while complying with platform rate limits. A structured anonymization process ensures privacy by hashing all identifiers, and a 24-dimensional feature vector is extracted for each post to capture sentiment, engagement, visual affect, and demographic indicators. Aggregated vectors serve as profile-level representations and enable scalable session simulations that estimate user exposure to hate. Using data from 80 Spanish-speaking influencers across six topical domains, the model accurately separates content types and produces hate-intensity scores that strongly correlate with manual labels. Session-level predictions have a mean absolute error within 4% of observed values, validating the utility of the vector model for rapid, privacy-preserving sentiment estimation. All code and a partially anonymized dataset are released to support reproducibility and are publicly available in the project repository.