About ReviewAid.

ReviewAid is an open-source AI-driven tool designed to streamline full-text screening and data extraction phases of systematic reviews. It leverages advanced large language models to classify papers based on PICO criteria and extract custom data fields, drastically reducing manual workload for researchers.

ReviewAid is not intended to replace manual screening and data extraction. Rather, it is designed to function as an independent supplementary reviewer, helping to minimize human error and enhance the overall precision, consistency, and reliability of the research process. Want more details / documentation about ReviewAid? Please visit aurumz-rgb.github.io/ReviewAid/ or Github Repository.

ReviewAid Data Extractor
ReviewAid Full text Screener
GitHub Repository Launch ReviewAid

If the primary ReviewAid streamlit is experiencing high usage, resource saturation, or memory limitations, users may utilize available mirror versions of ReviewAid without restriction.

ReviewAid Mirror

ReviewAid vs. Other Systematic Review Tools

ReviewAid uniquely combines full-text AI screening, customizable extraction, multi-provider LLM configuration, explicit confidence scoring, and local inference in one open-source tool.

Feature ReviewAid Rayyan Covidence ASReview
AI Full-text PDF Screening ✅ Yes ⚠️ AI-assisted ⚠️ AI-assisted ❌ Primarily title/abstract
Custom AI Data Extraction ✅ Yes ✅ Yes ✅ Yes ❌ No
Multi-LLM Support ✅ Yes ❌ No ❌ No ❌ No
Explicit Confidence Scoring ✅ Yes ❌ No equivalent system ❌ No equivalent system ❌ No equivalent system
Local / Offline AI ✅ Yes ❌ No ❌ No ✅ Yes
Free & Open Source ✅ Apache 2.0 ⚠️ Free tier / proprietary ❌ Proprietary ✅ MIT

Key Features

1. Full-text PICO Screening

AI-based PICO specified inclusion/exclusion classification

2. Full-text Data Extraction

Custom field extraction from text

3. Batch Processing

Process up to 20 papers at once

4. Multiple Exports

Export CSV, Excel, and Word formats

5. Live Terminal

Real-time processing logs to ensure transparency

6. Confidence Scoring

Estimates reliability of extraction to guide researchers to trust/not trust

7. Configuration

Configure any AI model using API key

8. Use

Runs locally/online, highly reusable

9. Open-source

Made by Researchers to ensure no proprietary "black box"

Confidence Scoring System

This layered approach ensures that high-confidence decisions are automated safely, while ambiguous or unreliable cases are clearly flagged for human oversight.

Confidence Score Classification Description Implication
0.9 – 1.0 Deterministic Match Extracted data was verified via exact string match or near-perfect token overlap against the source text. Safe to accept
0.6 – 0.89 High Confidence AI output verified via strong semantic token overlap. Paraphrasing detected and grounded in source. Review optional
0.4 – 0.59 Moderate Confidence Partial overlap detected. Ambiguous context or loosely met criteria. Manual verification recommended
0.1 – 0.39 Low Confidence Poor textual overlap or negation detected. AI score overridden by Tier 1 math due to likely hallucination. High risk of error
< 0.1 Unreliable Derived from fallback or failed extraction methods. Complete lack of text grounding. Mandatory manual review

Configuration

ReviewAid supports OpenAI GPT-4o for systematic review screening ReviewAid supports Anthropic Claude for AI data extraction ReviewAid supports DeepSeek for PICO screening ReviewAid supports Cohere for systematic review automation ReviewAid default AI model Z.ai GLM-4.6V-Flash ReviewAid supports Ollama for offline local systematic review screening

ReviewAid also supports configuration of OpenAI, Claude, Deepseek, Cohere, Z.ai and Ollama (locally) via API key as well. To protect your privacy, API keys are not stored at any time.

For tested tasks, the following models were successful:

OpenAI – GPT-4o

Deepseek – deepseek-chat

Cohere – command-a-03-2025

Z.AI – GLM-4.6V-Flash, GLM-4.5V-Flash

Anthropic – Claude-Sonnet-4-20250514

Ollama (local) – Llama3

Default – GLM-4.6V-Flash

Acknowledgements

ReviewAid powered by Z.ai GLM-4.6V-Flash AI model for systematic reviews

I gratefully acknowledge developers of GLM-4.6V-Flash (Z.ai) for providing the AI model used in ReviewAid.

The visual and text-based reasoning capabilities of GLM-4.6V-Flash have greatly enhanced ReviewAid's full-text screening and data extraction workflows.

For more information, please see GLM-4.6V-Flash paper and GLM-4.6V-Flash Hugging Face.

I would also like to thank Mohith Balakrishnan for his thorough validation of ReviewAid, including batch testing, error checks, and confidence verification, which significantly improved the tool's reliability and accuracy.

Citation

For ReviewAid's preprint paper, please check ReviewAid MetaArXiV.

If you use ReviewAid in your research, please cite it using the following format: