JULIE · Justified, Uniform Literature Integration Engine
A research instrument for evidence synthesis
Meet Julie
Julie is an AI research partner for systematic reviews, meta-analysis and evidence synthesis. She connects search planning, evidence assessment, pooling and manuscript drafting in a traceable workflow, and she works with both Oxford CEBM and GRADE. You set the direction and make the scientific decisions.
Julie is an AI research partner for systematic reviews, meta-analysis and evidence synthesis, working with both Oxford CEBM and GRADE. You set the direction and make the scientific decisions.
The bottlenecks in evidence synthesis (dual screening, data extraction, pooling, rating the evidence) are slow and expensive, and the shortcuts are indefensible in front of a reviewer. Julie is only worth using if the speed does not cost you the defensibility.
Nine stages, and six decisions that stay yours.
EXAMPLE · A GUIDELINE UPDATEOne review, imagined for this site. A guideline update with eleven questions, one methodologist, and an F2F consensus meeting already in the calendar. The methodologist is you, and each page picks up one moment of it.
The bottlenecks in evidence synthesis (dual screening, data extraction, pooling, rating the evidence) are slow and expensive, and the shortcuts are indefensible in front of a reviewer. Julie is only worth using if the speed does not cost you the defensibility.
Julie never gets tired of screening, and never makes one of your decisions for you.
Why it holds upTranscript
Every guideline rests on layers of evidence. In juvenile idiopathic arthritis alone, about two new papers appear every day.
In a classic review, two reviewers screen every paper by hand, and a third settles each disagreement. It is slow exactly where rigour matters most. In published medicine, about one quotation in six does not support its claim.
Our last review, done this way, took eighteen months, and its trail was hard to audit. This time, we asked Julie.
Julie keeps the rigour of a systematic review, and takes on the repetitive work. Two independent models screen every record; where they disagree, a person decides. Every number leads back to its source.
Built to bring it down to months, even weeks, with every step on record. At the consensus meeting, clinical experience decides. That part cannot be automated.
Julie does not run your searches, decide recommendations, or replace expert reading.
A general assistant answers. Julie follows the structure of a systematic review, and keeps a full audit trail. When a colleague asks where a number comes from, the answer is one line away.
In our own checks, trained reviewers made the same mistake as the model. A language model used without method is a trial run without randomisation. Neither is enough on its own. It takes a planned system.
Behind every line, a patient.
Sources: about two new JIA papers a day (730 with a 2025 publication date), our own PubMed count; about one quotation in six does not support its claim, Baethge and Jergas, Research Integrity and Peer Review, 2025. The article excerpt shown in the film is an illustration, not a real study.
Two ways to work with Julie
A systematic review, or a meta-analysis
-
Systematic review
From your questions to a manuscript draft: search strategies for three databases, every record screened twice, data extracted with the quote it rests on, evidence assessed with Oxford CEBM or GRADE, and the panel’s recommendations tied to the evidence behind them.
How it works -
Meta-analysis
When studies measure the same outcome in comparable ways, Julie pools them: from an analysis plan you sign, through forest plots and GRADE certainty, to a Methods section that names every decision. Start from a Julie systematic review, or from the studies you already have.
Meta-analysis
Systematic review · How it works
Nine stages, and where you decide
Julie follows the stages a systematic review actually has. What follows is that work from your side: what she prepares, what comes back to you, what you decide, and how the review carries on from your decision.
EXAMPLE · A GUIDELINE UPDATEIt crosses this chain once. The F2F consensus meeting you already have in the calendar is one of the decisions on the way, and it arrives with the panel's reading material already prepared.
- 1 research questions
- 2 eligibility criteria
- 3 search strategy preparation
- ↓ approve, run and export searches
- 4 title and abstract screening
- ↓ retrieving the full texts
- 5 full-text eligibility check
- 6 extraction and rating
- 7 material for the panel
- ↓ consensus meeting
- 8 linking evidence to recommendations
- 9 manuscript draft
The preparation starts with your question: Julie helps draft search strategies for PubMed, Embase and CENTRAL in the Cochrane Library. You review and approve them before running the searches. Each later decision stays with a named person, with a record of what was agreed.
How it works, stage by stageMeta-analysis · How it works
From the studies to a pooled estimate, and where you decide
Julie takes the studies, from her own review or from yours, through extraction by two readers, pooling and GRADE certainty to the Methods and Results. Four points wait for you, and each decision is recorded under your name.
Scope
Where Julie belongs, and where she does not
EXAMPLE · A GUIDELINE UPDATENothing in it happens at a bedside. It happens at a desk, months before anyone acts on it, and it has to hold up when a methods editor reads it. The distance is in time, not in purpose: there are patients behind every piece of evidence in it, and the recommendation is written for them.
Julie is
- A research instrument for evidence synthesis: systematic reviews, meta-analysis, guideline development, evidence summaries and health technology assessment.
- Built for work that will be read by a methods editor, a peer reviewer or a guideline panel, and has to survive that reading.
- A workbench with named human decision points, where the researcher decides and the machine does the volume work in between.
Julie is not
- Not an MD, and not just an .md either. She does not see patients or answer at the bedside, and she is more than a clever text file: she reads the literature behind the guideline and carries a review from the question to the manuscript, with every decision on the record.
- Not a clinical decision support tool. She answers nothing at the bedside, gives no treatment advice, and has no place in a consultation.
- Not a point-of-care reference. She is not an alternative to the summaries clinicians use to look something up during the working day.
- Database searches stay in your hands. Julie helps prepare the strategies for PubMed, Embase and CENTRAL. You approve them, run the searches and bring back the results.
- Not an autopilot. She does not decide what goes into your review, and she does not sign anything on your behalf.
The distinction matters because the two jobs have opposite failure modes. A point-of-care tool has to be fast and confident. An evidence synthesis instrument has to be slow enough to show its work, and honest enough to stop when it does not know something. None of that is academic caution. A recommendation that is wrong will be wrong in someone's treatment, and by then nobody is rereading the methods section.
Who Julie is forCapabilities
What you get
Systematic review
Every record is assessed twice against your criteria: where the two assessments agree the decision stands, and where they differ you decide.
Dual screening with human adjudication, stated in your methods section, even when there is no second person in the building.
Evidence rated by Oxford CEBM or by GRADE as published, chosen for each project, with every rating traceable to the quotations it rests on.
Every human decision under the name of the person who made it, so the review can be reported and repeated by someone who was not there.
Meta-analysis
An analysis plan you sign before any pooled number exists, with every later change reported as one.
Pooled estimates calculated by established meta-analysis software, not by the AI model, so anyone can recompute them.
Every contributing number with a recorded decision behind it, and every study left out with its reason.
On GRADE, a Summary of Findings that rests on the same numbers as the forest plot.
Already using Consensus, Elicit or SciSpace?
The question is what your project needs next. Julie supports a researcher-led workflow that connects search planning, evidence assessment, agreed recommendations and a Discussion draft.
For teams preparing a systematic review or guideline update, that connection matters: the manuscript should reflect the decisions you actually made, with the supporting and contradicting evidence still traceable.
Contact
Get in touch
Tell us a little about your review or panel. A person will reply.
EXAMPLE · A GUIDELINE UPDATEThis one was imagined for the site. Yours is not, so tell us what yours looks like.