← allmeta

RCT Extractor DEMO

Extract effect estimates (HR, OR, RR, MD, SMD, IRR, RD) from RCT publications. Works fully offline in the browser (a JS port of the rct-extractor-v2 patterns), or against your locally-running rct-extractor-v2 API for the full 180+ pattern engine. Optional: run a local LLM alongside as a cross-check. Consensus and disagreements are highlighted.

checking rct-extractor… checking local LLM…

Input

First time? In a terminal: cd C:\Projects\rct-extractor-v2 && pip install fastapi uvicorn python-multipart && uvicorn src.api.main:app --port 8000. Then come back and click Extract.
Demo mode active. Results are sample data, not real extractions.
Import ClinicalTrials.gov results (JSON, offline)

Download a study's results from clinicaltrials.gov/api/v2/studies/<NCT>?format=json, then paste or upload it here. Parsed entirely offline — no network call is made.

Extractions

SourceTypeEstimate95% CIpConf
No extractions yet.

Consensus vs local LLM

Enable "local LLM consensus" above to compare the regex extractor against a local model. Disagreements are highlighted red.

Source provenance

Horizontal bar chart: one row per extraction. Each bar's length and color encode the regex extractor's confidence (green ≥ 0.8, amber ≥ 0.5, red otherwise). A per-row title on hover gives the precise confidence and tier.