AI-assisted systematic review software: From papers to reviewed evidence.

eviStreams is AI-assisted software for systematic reviews. AI pulls each value from the paper and marks the sentence it came from, two reviewers check it, and you get evidence you can defend.

PDFRISSMITH ET AL. · 20244Methods2.1 Participants and randomisationWe enrolled 124 patients aged 3–6 yearsfrom four community clinics and randomisedthem to 38% SDF or placebo varnish.Allocation was concealed in sequentiallynumbered opaque envelopes prepared by anindependent statistician. Children wereeligible if they had at least one activelesion in a primary tooth.2.2 OutcomesThe primary outcome was caries arrest at12 months, assessed by two calibratedexaminers blinded to allocation.Secondary outcomes were analysed post hoc.RoB 2ASSESSEDSMITH 2024R1 ✓ R2 ✓1243–6 ySDF 38%AI marksthe exactsentencePDFs, RIS,PubMed inR1 + R2 check every valuepooled into a forest plot,exports: CSV, JSON, SVGrisk of bias,judged with RoB 2
  • AI marks the exact sentence
  • PDFs, RIS, PubMed in
  • R1 + R2 check every value
  • Risk of bias with RoB 2
  • Pooled into a forest plot
  • Exports: CSV, JSON, SVG
Example · Smith 2024, page 4
Built at
University of PennsylvaniaDeveloped within Penn Dental Medicine
02 Review and consensus

Independent reviews. One recorded decision.

Reviewers enter values on their own. Where they differ, the consensus reviewer sees every entry with its evidence, a majority suggestion when two agree, and records one decision. Try it: accept the majority.

  1. 1R1 says 12, R2 says 18
  2. 2every answer keeps its quote and page
  3. 3one click, one recorded decision
Consensus · Smith 2024
24 fields · 18 agreed · 3 resolved · 3 need decision · consensus reviewer CR
Example data
sample_size124AI · R1 · R2 matchAgreed
follow_up_monthsNeeds decision
Months from randomisation to the primary outcome · number
AI AI
12
“The primary outcome was assessed at 12 months” p.7
REVIEWER 1 R1
12
“The primary outcome was assessed at 12 months” p.7
REVIEWER 2 R2
18
“Table 2. Caries arrest at 18 months follow-up” p.9
AI + R1 agree (2 of 3) · 12
Or pick a source box above, or Quotes open the PDF at the cited page.
follow_up_months12CR accepted the majority · AI + R1Decided
caries_arrest_pct68.4 · 68.4 · 71.0Needs decision
03 Risk of bias

Assess risk of bias with the evidence in view.

Reviewers work through the RoB 2 domains for parallel-group, cluster-randomized and crossover trials with the supporting quotes visible, and settle differences into one recorded assessment.

  1. 1D5: the two readers disagree
  2. 2the quote behind the judgement
Risk of bias · Smith 2024RoB 2 · parallel-group RCTExample data
DomainR1R2Consensus
D1 · Randomization processLowLowLow
D2 · Deviations from interventionLowSome concernsLow
D3 · Missing outcome dataSome concernsSome concernsSome concerns
D4 · Measurement of the outcomeLowLowLow
D5 · Selection of the reported resultSome concernsHighPending
OverallSome concernsHighPending
Secondary outcomes were analysed post hocD5 · R2 · p. 9
LowSome concernsHighPending
04 Synthesis

Reviewed evidence becomes synthesis.

Map fields to outcomes and comparisons, confirm eligible studies, then pool with a forest plot and diagnostics, or synthesize without pooling when pooling is not appropriate.

1 · Reviewed evidence5 studies
Smith 202448/62 · 30/60 · consensusEligible
Johnson 202370/95 · 44/92 · consensusEligible
Chen 202322/40 · 18/41 · NaF comparatorExcluded
Patel 202555/80 · 35/78 · consensusEligible
Okafor 202215/30 · 12/31 · figure-verifiedEligible
2 · Eligibility and mapping
  • caries_arrest → events · sample_size → N
  • Comparison: SDF 38% vs placebo
  • Chen 2023 excluded: comparator is NaF, not placebo
  • Okafor 2022 events read from Figure 2, reviewer-verified
3 · Synthesis✓ Map fields✓ Choose comparisonPlotDiagnosticsExample data
Caries arrest · 12 moParent-reported pain · 12 mo
Caries arrest at 12 monthsSDF 38% vs placebo · risk ratio · random effects (DL)Export CSVSVG
STUDYSDF N/NPLAC. N/NRR (LOG SCALE)RR [95% CI]WEIGHTSmith 202448/6230/601.55 [1.16, 2.06]27.6%Johnson 202370/9544/921.54 [1.21, 1.97]37.8%Patel 202555/8035/781.53 [1.15, 2.04]27.6%Okafor 202215/3012/311.29 [0.73, 2.28]7.0%Total (95% CI)188/267121/2611.52 [1.31, 1.77]100%0.5124← favours placebofavours SDF →
τ² = 0.00 · Q = 0.34, df = 3 (p = 0.95) · I² = 0% Low heterogeneityZ = 5.46 (p < 0.001)
  1. 1Chen 2023 stays out: wrong comparator
  2. 2pooled, with its diagnostics
  3. 3export the figure and the data behind it
05 Team coordination

One workspace for the review team.

Project owners define the scope and assign reviewer seats. Each reviewer sees the studies waiting on them; leads see progress across extraction, consensus and appraisal.

SDF
Silver diamine fluoride review
48 studies · 3 reviewer seats · owner: project lead
OverviewDocumentsFormsConsensusRisk of biasSynthesis
Example data
Collect 48/48
Extract 44/48
Review 31/48
Assess 18/48
Synthesize –
Review scopeEdit
PChildren 3–6 with active cariesISilver diamine fluoride 38%CPlacebo varnishOCaries arrest at 12 months
patient_population_form 24 fields · RoB 2 enabled
Assignments Manage seats
R1Reviewer 1extraction · RoB 2 · 24 studies
R2Reviewer 2extraction · RoB 2 · 24 studies
CRConsensus reviewerall fields
Two independent extractions per study; one consensus reviewer records decisions.
My queue · R13 waiting
Chen 2023 Manual extractOpen →
Patel 2025 RoB 2 · D3–D5Open →
Okafor 2022 Verify figureOpen →
Only the studies assigned to you, in the order they were queued.
Jobs Last run 14:12
Extraction · batch 316/16Done
Extraction · batch 414/162 failed
RoB 2 · signalling questions18/48Running
Failed items keep their inputs and can be retried in place.
  1. 1scope set once, as PICO
  2. 2each reviewer sees only their own queue
  3. 3failed jobs retry in place
06 Inputs and outputs

Bring evidence in. Take reviewed work out.

Records arrive from the sources review teams already use. What leaves carries the review with it: values with their source pages, judgements with their evidence, and plots with the study-level data behind them.

In Documents
PDFFull-text papersupload or drag in
RISReference librariesRIS and EndNote
DOIIdentifiersresolved to records
DBSearchPubMed, ClinicalTrials.gov
Silver diamine fluoride review 48 studies
Collect & configure48/48
Extract44/48
Review31/48
Assess · RoB 218/48
Synthesize–
Everything on the right comes from this one project.
OutExport
Extraction resultsEvery field per study, with source page and quote columns when you want themCSVJSON
Risk-of-bias tableDomain judgements for each study, plus a per-study reportCSVReport
Forest plot and study dataPublication-ready figure with the study-level table behind itSVGPNGJPGCSV
  1. 1PubMed search built in
  2. 2exports keep the link to the reviewed record
Security

Built for research data.

How eviStreams handles research data
  • Access by projectPeople see only the projects they belong to, and their role decides what they can change.
  • Blinded reviewersSwitch it on and the two reviewers cannot see each other’s answers until both have finished.
  • Short-lived links to papersPDFs open through signed links that expire after an hour, never a permanent public address.
  • Every decision on recordReviewer answers and consensus decisions are kept with who made them and when.

Questions teams ask first.

Q1

What is eviStreams?

eviStreams is AI-assisted systematic review software for research teams. It brings study collection, structured data extraction, independent review, reviewer consensus, risk-of-bias assessment, and evidence synthesis into one workspace. AI proposes answers; your team checks the evidence and records the decisions.

built at Penn Dental Medicine
Q2

Who is eviStreams for?

Research teams conducting systematic reviews and evidence synthesis. The workspace connects study collection, AI-assisted extraction, human review, risk-of-bias assessment, and synthesis.

Q3

Where does AI fit into the review?

AI can assist with extraction and suggestions. Reviewers inspect the evidence and make the recorded decisions.

Q4

What can I import?

PDFs, RIS and EndNote files, DOIs, and records found through PubMed and ClinicalTrials.gov search.

Q5

Can I check an extracted answer against the paper?

Yes. Reviewers can inspect the available supporting passage, table, figure, or page context alongside an extracted value. Source references still need to be checked for the correct study arm, outcome, and time point.

Q6

How are reviewer disagreements resolved?

An adjudicator compares the independent entries with their evidence and records a consensus decision. A majority suggestion helps organize review; the reviewer remains responsible for the decision.

Q7

Which risk-of-bias tool is supported?

RoB 2 for randomized trials: parallel-group, cluster-randomized and crossover.

Q8

Do I have to pool studies?

No. When pooling is not appropriate, eviStreams can retain the reported results and their sources for structured synthesis without a pooled estimate.

Q9

Can I try the workflow before registering?

Yes. The public demo introduces a private temporary workspace with example studies and explains its usage limits before you start.

Read the systematic review workflow guides

See it on your own papers.

Request access for your team, or open the demo workspace with example studies first.