RCT-Reviewer 3.0

v3.0 represents a significant step forward.

Major improvements include:

• Polished evidence PDF exports
• Improved highlighting engine
• Better PDF formatting and pagination
• Improved Hugging Face model handling
• Improved Streamlit integration
• Better logging and debugging
• Numerous bug fixes and UI refinements

RCT-Reviewer v2.0

• 3 prototype .py runners to run RobotReviewer ML models from 3 different sources
• Improved Pipeline

RCT-Reviewer v1.0

RCT-Reviewer v1.0 is a modernized machine learning–based system for analyzing Randomized Controlled Trials (RCTs).

v1.0 is a prototype version that runs fully locally using pretrained models stored and accessed via Git LFS.

• Predicts risk-of-bias across standard domains using RobotReviewer models
• Accepts RCT PDFs and extracts structured text
• Applies NLP preprocessing and section segmentation
• Uses locally stored pretrained classical ML models for inference
• Lightweight design optimized for low-resource environments

Relation to RobotReviewer:
RCT-Reviewer is a modernization of RobotReviewer (automated RCT analysis and risk-of-bias prediction). It does not introduce a new scientific method but reimplements and restructures the original pipeline into a more modern, modular system.