Our Methodology
A systematic approach combining 75 years of the Delphi method with multi-model AI consensus — transparent, reproducible, and built for researchers.
The Delphi Method
The Delphi method was developed at the RAND Corporation in 1963 by Dalkey and Helmer. For over 75 years, it has been used in more than 10,000 published studies to reach reliable consensus among independent experts.
Delph-AI adapts this proven methodology to AI-powered systematic review screening. We brought large language models into the expert panel — same rigor, faster results, with you always in control.
published studies
Dalkey & Helmer, RAND Corporation, 1963
The Screening Pipeline
A six-step process that mirrors best practices in systematic review methodology
Data Ingestion
Import references from PubMed, Scopus, Web of Science, or any database. Automatic deduplication across sources.
Criteria Definition
Set inclusion/categorization and exclusion criteria in plain language. AI assistant helps refine them using PICO frameworks.
Independent Evaluation
Multiple AI models evaluate each record against each criterion independently — then perform a consensus panel for each disagreement.
Consensus Calculation
Agreement rate quantifies confidence across all models. Disagreements surface uncertain cases for human review.
Shortlist Generation
Get your evidence shortlist ranked by agreement rate with full audit trail — every model, every vote, every reason.
Version Control
Change which criteria are exclusion criteria to generate new shortlists without re-running the screening. Each version is immutable, comparable, and auditable.
Transparency at Every Step
Every screening includes full visibility into the AI decision process — compliant with EU AI Act Article 50
25 AI Models, 8 Providers
From GPT-4o to Gemini Pro to Claude — each model evaluates independently. You see which models reviewed each record, how they voted, and why.
Agreement Rate
Confidence is quantified, not assumed. The agreement rate across all selected models determines inclusion — no single model decides alone.
Full Reproducibility
Same data + same criteria + same models = same consensus for your results. Every version is immutable and auditable. No black boxes.
The Human Always Decides
AI prioritizes and justifies. The researcher has the last word. Delph-AI is a tool that augments your judgment — it never replaces it.
See it in action
Start your first screening today. Define your criteria, choose your AI panel, and review the consensus.
Sign up for free and screen up to 500 records with our trial.