Open tools for medical evidence synthesis
Open infrastructure for medical evidence synthesis.
Medevi leads the development of open-source tools and public benchmarks for systematic reviews: screening, data extraction and critical appraisal with large language models, where every judgement traces back to its source.
- peer-reviewed studies in our MERSQI and NOS-E appraisal benchmark
- 139
- outcome-level RoB 2 judgements across 154 RCTs in Medevi-RoB
- 273
- ICC with human reviewers on total MERSQI score (human–human: 0.79)
- 0.755
- per paper, against 65–130 minutes of manual extraction
- 29–57 s
Why Medevi
A systematic review costs $50,000–$140,000 and takes 6–18 months. Risk-of-bias assessment alone takes 6–12 weeks. AI can carry part of that load only if it is open, measured against human experts, and auditable line by line.
Closed tools are hard to validate and consumer chatbots hallucinate. Medevi publishes its pipelines, prompts and benchmarks so that reviewers, journals and guideline bodies can check the method for themselves.
Platform
One pipeline, from PDF to appraisal.
Bring your own model and endpoint. Medevi handles the steps between a manuscript and a defensible judgement.
- 01
Preprocess
OCR and text processing of manuscripts, combined with your questions and the operational definitions of the appraisal tool.
- 02
Extract
Single-shot extraction for speed, or per-domain and rules-based modes for precision on tools like RoB 2.
- 03
Validate
Uncertainty quantification with black-box, white-box and LLM-as-judge methods flags the calls a human should review.
- 04
Report
Structured output, the model's rationale and supporting evidence, exportable for your review software.
Example output
Every judgement ships with its score, certainty and the passage it came from.
- certainty
- 0.96
- score
- 1.5 / 3
- mode
- per-domain chain-of-thought
- evidence
- “…compare student cohorts' performance with or without retrieval practice in renal physiology classes, relative to the national average…”
Open benchmarks
We measure models against expert reviewers, and publish where they fall short.
Systematic reviews & meta-analyses
Every stage of the PRISMA flow, with reviewers in control.
Medevi follows the structure of a PRISMA 2020 review. AI assists at each stage, from de-duplication and screening to data extraction and risk-of-bias appraisal, and every inclusion decision stays with your review team.
Open, evidence-based risk scores for clinical medicine.
Ready for MCP usage and integration into EMRs.
We are bringing the same open, auditable approach from evidence synthesis to the bedside: open-source calculation of validated clinical risk scores such as CHA₂DS₂-VASc, Wells' Criteria for PE and the NIHSS, each linked to the evidence behind it.
Interested in piloting? Email us- Open sourceEvery formula and implementation is public and auditable.
- Evidence-linkedEach score cites the derivation and validation studies behind it.
- Ready for MCP and EMRsExposed as Model Context Protocol (MCP) tools for AI assistants, and built for integration into electronic medical records.
Latest
All news →Made possible by supporters at:
Run your next review on open infrastructure.
Medevi is in private alpha with research teams. Email us to request access and we'll be in touch.