Moein Taherinezhad, Sebastian Maier, Gerardo Vitagliano +2 more
A meta-analysis is one of the more grinding jobs in research: track down every relevant study, screen them, read the full texts, pull out the right numbers, convert them to a common scale, and pool them into a single estimate. AutoSynthesis automates that whole assembly line with a team of agents. You hand it a research question in plain language, and it runs the pipeline end to end, from building a search strategy to a random-effects pooled result.
It also does some of the quality work a careful reviewer would, like checking how effects vary across study features and flagging risk of bias, then writing up a report following PRISMA reporting standards. In their trial it screened over 28 studies and extracted more than 20 quantitative claims, landing pooled estimates close to what human experts got.
That agreement is encouraging but comes from the abstract on a single application, so the paper is the place to judge how well it holds up across messier evidence bases.
Evidence synthesis is crucial for turning primary research into reliable knowledge for science, medicine, education, and policy. Yet, quantitative evidence synthesis remains largely manual and difficult to scale. Here, we introduce AutoSynthesis, an end-to-end multi-agent system for automated meta-analysis. Given a research question in natural language, AutoSynthesis formulates a search strategy, retrieves scientific literature, screens candidate studies, assesses full-text…
Loop the Loopies!
arXiv (cs.LG) · July 17, 2026DELUGE: Towards Continental-Scale Daily Pluvial Flood Damage Prediction via Interpretable Conditioning on Foundation Model Embeddings
arXiv (cs.AI) · July 17, 2026SciForge: An AI-Native, Multimodal Workbench for Scientific Discovery
arXiv (cs.AI) · July 17, 2026Revisiting data-driven dynamic security assessment with a tabular foundation model