MetaScreener automates the screening phase of systematic reviews by running multiple open-source LLMs in parallel and aggregating their outputs through a calibrated confidence pipeline. Users upload search results from PubMed/Scopus, define review criteria using PICO/PEO/SPIDER frameworks, and receive include/exclude decisions with confidence scores. The system routes uncertain cases to human reviewers while making automatic decisions on high-confidence cases, incorporating active learning to recalibrate model weights based on human feedback.