Abstract
Systematic reviews are widely utilized as a rigorous method for synthesizing scientific evidence, yet the manual process of screening literature and extracting data remains a time-consuming bottleneck. Researchers often spend weeks sifting through thousands of articles to identify relevant data. This paper presents ReviewAid (v2.1.0), an open-source, AI-powered software tool designed to automate and accelerate this workflow. Unlike single-model solutions, ReviewAid offers a flexible architecture supporting multiple AI providers, including OpenAI, Anthropic, DeepSeek, Cohere, Z.ai, and local execution via Ollama, allowing researchers to balance cost, speed, and privacy. Built on Python and Streamlit, ReviewAid facilitates “PICO-based” screening, a structured method used in healthcare to identify studies based on Population, Intervention, Comparison, and Outcome, as well as customizable data extraction. A significant challenge in using AI for research is that models often produce “malformed” outputs, such as text with broken formatting, which can cause standard software to crash. To address this, ReviewAid introduces a ‘Bulletproof Parsing Pipeline’ designed to recover data from these imperfect responses. Additionally, it features a hierarchical four-tier confidence scoring system to quantify the certainty of AI decisions. Software validation on over 100 articles demonstrated high processing speeds and robust error handling. ReviewAid is architected as a decision-support tool, not as a replacement for human judgment, but as a ‘third reference’ layer to assist the review process, distributed under an Apache 2.0 license.