A research team has published the architecture for meta-pipe, an open-source AI pipeline that handles the full systematic review and meta-analysis workflow from end to end.
The tool chains Claude models — Opus 4 for reasoning, Haiku 3.5 for cheaper classification tasks — with roughly 3,600 lines of Python automation, R statistical packages, and Quarto for document rendering. The pipeline runs across 10 stages and costs an estimated $15-30 per review covering five to ten studies. Four capabilities the authors say no single competing tool currently offers: automated manuscript generation, semi-automated GRADE evidence quality scoring, overclaim detection across 12 predefined patterns, and dual-paradigm network meta-analysis running both Bayesian and frequentist approaches. Five human decision points are baked in as mandatory checkpoints, not optional ones.
Systematic reviews are slow, expensive, and bottleneck evidence-based medicine — a credible automation layer could meaningfully accelerate how clinical guidelines get updated. The overclaim detection feature is the most interesting angle: building a skepticism check directly into a pipeline designed to generate manuscripts is a rare piece of epistemic self-awareness in AI tooling.
The authors are direct about the obvious gap: there is no validation data. meta-pipe is a system description, not a proven tool, and formal testing against published Cochrane reviews is still underway. A $20 run through a pipeline that hallucinates a confidence interval is not a bargain.