EvidenceEngine: AI-Assisted Systematic Review Pipeline for Data Extraction

EvidenceEngine provides a standardized, reproducible Human-in-the-Loop pipeline for acting as an Automated Independent Second Reviewer.

While it utilizes frontier Large Language Models (LLMs) to perform expert data extraction, Risk of Bias audits, and Results Synthesis, the system is designed to augment—not replace—human expertise.

By generating side-by-side reconciliation logs for every AI decision, the engine ensures 100% manual verification of all findings.

This hybrid approach guarantees rigorous scientific transparency and maintains "Meaningful Human Control" over the final evidence synthesis, meeting the highest 2026 standards for AI-assisted systematic reviews.

COCHRANE & METHODOLOGICAL COMPLIANCE

This engine is architected to satisfy the MECIR (Methodological Expectations of Cochrane Intervention Reviews) standards:

• Standard C43 (Structured Forms): The 'promptfile.txt' and 'criteria.txt' serve as digital standardized data collection forms, ensuring unbiased and consistent extraction across all studies.
   
• Standards C45/C46 (Independent Dual Processing): The script acts as the Independent Second Reviewer. The reconciliation workflow allows the primary researcher to independently verify AI decisions, fulfilling the requirement for dual-independent screening and extraction.
• Standard C44 (Detail Requirements): Pre-configured prompts ensure the capture of mandatory "Characteristics of Included Studies" (PICO, settings, study design, and Risk of Bias) required for Cochrane Evidence Tables.
• PRISMA-trAIce (2026): Every run generates a technical log (timestamp, model version, and full prompt), ensuring the AI's "logic" is fully auditable and transparent for peer review.