Data Extraction for Systematic Reviews
Still typing numbers out of PDFs
into a spreadsheet?
Read the full text, ground every value to its source quote, and hand you a reviewable table — in minutes, not weeks. Free, restricted demo, no setup required.
Try the live demo
Define a research question, upload up to three of your own full-text PDFs (or use the built-in sample paper), and see a grounded extraction table with confidence and evidence. Leave your work email and you're in.
Three screens, two decisions
The full production tool has eight steps and an append-only audit trail. The demo keeps the two decisions that matter to you and runs everything else — parsing, segmentation, protocol confirmation — automatically behind the scenes.
1. Define
Describe your review question in plain language. The system derives the PICO framing and suggests a starting set of extraction fields for you to review and approve.
2. Upload
Drop up to three full-text PDFs — or skip straight to a precomputed sample paper to see a result with zero setup.
3. Extract
Watch the pipeline run through parsing, structuring, and extraction, then review a results table with confidence pills and expandable evidence quotes.
Provenance is the point, not an afterthought
Automating the reading is only useful if you can still trust and check every number. That is the design constraint the whole tool is built around.
Evidence travels with the value
Every extracted number carries the source quote, section, and page it came from — expand any cell to see exactly where it was found.
Confidence, not one blended number
Presence, grounding, and domain validation are reported and scored separately, so you can see which one is uncertain instead of trusting a single percentage.
Deterministic where it can be
Parsing, sectioning, and bibliographic metadata are resolved with deterministic code. The language model is only asked to read — not to invent.
Ephemeral by design
Uploaded PDFs are processed and deleted. Nothing from the demo is used to train anything, and demo projects expire automatically.
EU data hosting
Operated on secure servers in Germany, in line with the rest of MH-Analytics infrastructure.
Full audit trail in production
The production tool binds every exported value to a protocol hash, segmentation snapshot, model, and run ID. The demo shows the review surface; the export/manifest step is disabled here.
Read the full case for auditable extraction
Want the reasoning, the accuracy numbers, and what to demand from any extraction tool? Read the full write-up, or reach out for a walkthrough on your own review.
Questions? Call +49 1515 79 43 500 or write to marc.harms@mh-analytics.eu.