Meta-Mar: An AI-Assisted Web Platform for Accessible and Rigorous Meta-Analysis

Meta-analysis underpins evidence-based research by synthesizing findings across studies to produce more reliable and generalizable conclusions. Yet existing tools remain fragmented—often technically demanding, costly, or lacking transparency—creating barriers for early-career researchers and interdisciplinary teams. This thesis addresses that challenge through the design and evaluation of Meta-Mar, an AI-assisted, web-based meta-analysis platform that combines statistical rigor with accessibility and ethical data governance. Developed in R using the meta and metafor packages, Meta-Mar performs fixed- and random-effects meta-analyses across continuous, binary, correlation, and generic data structures. It integrates advanced heterogeneity diagnostics, publication-bias assessments, and publication-grade visualizations within a guided, stepwise interface. The inclusion of large language models adds contextual assistance for model selection, data validation, and interpretive reporting—bridging the gap between statistical complexity and practical usability for researchers unfamiliar with meta-analytic conventions. A comparative validation against RevMan 5 demonstrates computational parity in effect-size estimation, confidence intervals, and study weighting, confirming analytical reliability. The platform incorporates a privacy-by-design framework aligned with GDPR principles, employing automated anonymization, encryption, and explicit user consent mechanisms. With over 5,800 users and adoption in graduate-level workshops, Meta-Mar demonstrates real-world viability alongside technical soundness. Beyond its technical implementation, this work contributes to a broader vision of AI-supported evidence synthesis—enhancing scientific reproducibility, reducing dependence on proprietary tools, and facilitating data-driven decision-making across health, education, and behavioral science. Future developments will extend statistical capabilities, refine AI explainability, and introduce publication-ready reporting templates, positioning Meta-Mar as an evolving open resource for accessible meta-analytic research.Keywords: meta-analysis, evidence synthesis, AI-assisted research, reproducible analytics, heterogeneity assessment, publication bias, Meta-Mar