AIM Review

Home

ARTIFICIAL INTELLIGENCE APPLIED TO SYSTEMATIC REVIEWS

AIM REVIEW FLASH

Streamline your systematic reviews with benchmarked and trusted AI algorithms.

AIM Review is an integrated research platform for the literature-review workflow. The platform provides AI-assisted tools to support the entire review process: from title and abstract screening to AI-assisted full-text structured data extraction.

Use AI and machine-learning methods to reduce repetitive review work while keeping researchers in control of the decisions that shape their evidence base.

Video tutorials >> Step-by-step tutorials >> Documentation >>

Supported by King's College London

Manage your projects.

Create, share, or delete cloud projects

Projects

0 of 2 owned cloud projects

Loading projects

Connecting to your AIM Review account.

Project setup

Load references into a project

Import Excel, CSV, RIS, PubMed/NBIB, or labelling snapshot files. The same parser used by the pro web app cleans empty entries and duplicates before saving.

Owned projects

Choose where to load

Screening

Label project records

Open an owned or shared project, label the selected records, write notes, and save progress back to the same cloud project.

Projects

Open for labelling

Full text review

Extract data from PDFs

Load references, attach local PDFs, parse the full text, and answer extraction questions manually or with a configured LLM service.

Local workspaces

Open data extraction

Review agreement

Collect and resolve shared screenings

For owned cloud projects with accepted collaborators, collect saved screenings, inspect agreement, and resolve conflicting records.

Owned projects

Build agreement

Automatic screening

ML prediction

Projects

Choose labelled data

Application settings

Configuration

Labelling configuration

Display

1x
Dark mode

Text highlights

Custom

ML configuration

Active learning

Active learning and ML prediction

Help and workflow

Documentation

Flash guide

AIM Review Flash

AIM Review Flash is the streamlined version of AIM Review. It focuses on the core review workflow using default parameters and with minimal configuration options. You can create or open a project, load references, screen records, extract data from local PDFs, check agreement, and use local machine learning to prioritise relevant publications.

AIM lab Publication

Learn by doing

Interactive tutorials

Want to learn by doing? The interactive tutorials are a hands-on, step-by-step companion that lives inside the app — a quicker way to pick up the whole Flash workflow than trawling through this guide. Open them, follow the steps beside the sections, and tick each one off as you go — your progress is remembered. You can hop over to any section at any time and press Documentation to come straight back to where you left off.

The screening workflow

1
Prepare the review question

Write the research question, eligibility criteria, and the decision rules that define relevant and irrelevant records before importing data.

2
Search bibliographic databases

Export records from sources such as PubMed, Web of Science, Scopus, PsycINFO, or other databases. Keep the search strategy and export date for reporting.

3
Load records into AIM Review

Use Load Data to import database exports into a cloud project, or use the local file option for a no-login workflow.

4
Screen titles and abstracts

Use Labelling to mark each record as relevant or irrelevant. Save progress regularly and use notes for borderline decisions.

5
Check agreement or train prediction

For multiple raters, use Agreement to resolve conflicts. For larger reviews, use ML Prediction after enough examples have been labelled. ML Prediction uses supervised learning to predict the labels of unscreened publications.

6
Extract full-text data

Use Data extraction to attach PDFs, parse full text, define extraction questions, and save local JSON snapshots for reopening progress.

7
Export and report

Export labelled records, extraction tables, agreement results, or snapshots. Document the software version, access date, labels used for modelling, and decision thresholds.

Profile

Profile is where signed-in users manage cloud projects. A project is the container for references, labels, sharing, and saved progress.

  • Create a cloud project before loading records.
  • Use clear project names that identify the review and screening stage.
  • Share projects when more than one reviewer will screen the same records.
  • Local file workflows do not require Profile and remain in the browser tab until exported.

Load Data

Load Data imports bibliographic records and prepares them for screening. Use this first when starting from database exports.

  • Supported inputs include Excel, CSV, RIS, PubMed text or NBIB, and AIM Review JSON snapshots.
  • At minimum, each record should include a title, abstract, and stable identifier such as DOI, PMID, or accession number.
  • AIM Review removes repeated and incomplete entries where possible and reports loaded, duplicate, and incomplete counts.
  • For cloud work, choose a project. For no-login work, use the compact file option at the bottom of the Load Data screen.

Labelling

Labelling is the main screening workspace. It presents one record at a time so reviewers can make consistent inclusion decisions.

  • Mark records as relevant or irrelevant using the buttons, swipe actions, or keyboard shortcuts.
  • Use notes for uncertainty, missing information, or decisions that need full-text checking.
  • Use active learning after at least some relevant and irrelevant examples have been labelled.
  • Export .xlsx for spreadsheet review and .json when you want to reopen the same progress in AIM Review.

Data Extraction

Data Extraction is a local full-text workspace for PDFs and structured extraction questions.

  • Open an Excel reference table or JSON extraction snapshot, then attach PDFs in the workspace.
  • Attach PDFs locally, parse their text in the browser, and answer questions manually when needed.
  • Use Unmatched PDFs: N to review PDFs that were not found in the reference list. The review panel keeps controls on the left and the PDF on the right, and lets you match the PDF to a reference without adding it to the main table by mistake.
  • Use the compact LLM configuration to choose an API service, model, and task. Automatic extraction fills unanswered fields, while cross-checking reviews answers already present in the table and records whether they are supported, conflicting, or unclear.
  • Save a JSON snapshot to reopen progress later. The snapshot keeps records, questions, answers, cross-check metadata, PDF metadata, unmatched PDF metadata, and parsed text.

Agreement

Agreement helps owners combine screening decisions from multiple reviewers and resolve disagreements.

  • Use cloud sharing when reviewers need to screen the same project independently.
  • Collect screenings after reviewers have saved their labels.
  • Inspect rater labels, resolve conflicts, and optionally apply labels where reviewers agree.
  • Export agreement results to preserve resolved decisions and agreement statistics.

ML Prediction

ML Prediction trains a lightweight local model from labelled examples and previews automatic labels for unlabelled records.

  • Label at least 4 relevant and 4 irrelevant records before training.
  • Review model performance before applying automatic labels.
  • Choose a threshold or confidence rule based on the review's tolerance for missing relevant records.
  • Automatic labels are staged first, so the project is not changed until you save them.

Configuration

Configuration controls how records are displayed and highlighted during screening.

  • Adjust font size and dark mode for comfortable long screening sessions.
  • Choose which metadata fields are shown on each record.
  • Add relevant, irrelevant, or custom highlight expressions to make key terms stand out.
  • Use ML Configuration to choose the local machine-learning method used by active learning and prediction.

Cloud project or local files?

Cloud projects

Best when you want saved progress across devices, project sharing, or agreement collection from multiple reviewers.

  • Requires sign-in.
  • Saves progress to the selected project.
  • Supports sharing and agreement collection.
Local files

Best for a quick no-login workflow, offline-style work in one browser tab, or continuing from exported .xlsx and .json files.

  • No sign-in required.
  • Export .xlsx or .json to keep progress.
  • Local sessions disappear when the tab is closed unless exported.

Practical screening guidance

Decide the rules before screening. AIM Review can speed up the workflow, but it does not replace review judgement, protocol decisions, or final eligibility checks.

  • Start with clear criteria. Define population, intervention or exposure, comparator, outcomes, study design, date limits, and language limits where relevant.
  • Pilot the first records. Screen a small batch and refine the decision rules before reviewing the full dataset.
  • Use notes consistently. Notes are useful for uncertainty, full-text retrieval, duplicate concerns, or reasons for exclusion.
  • Monitor class balance. Active learning and prediction need both relevant and irrelevant examples. If all early labels are one class, keep manual screening until both classes are present.
  • Do not blindly accept predictions. Treat ML outputs as assistance. Check model performance, thresholds, and a sample of automatic labels before relying on them.

Reporting

Methods sections should make the workflow reproducible. Record what was done in the databases, in AIM Review, and during any manual checks.

Minimum software details
  • Software name: AIM Review Flash.
  • Date accessed or date used.
  • Whether cloud projects or local files were used.
  • Input fields used for screening, usually title and abstract.
If machine learning was used
  • Number of manually labelled examples.
  • Selected ML method from ML Configuration.
  • Validation criterion and observed performance metrics.
  • Threshold or confidence rule used to stage automatic labels.
If agreement was used
  • Number of reviewers.
  • How disagreements were resolved.
  • Whether agreed or majority labels were applied.
  • Final number of resolved relevant and irrelevant records.

Troubleshooting

My file does not load

Check that it is one of the supported file types and that title, abstract, and identifier columns are present. For saved progress, use AIM Review .xlsx or .json exports.

I cannot see cloud projects

Make sure you are signed in and refresh the project list. Local file work remains available without signing in.

ML Prediction says more labels are needed

The Flash prediction workflow requires at least 4 relevant and 4 irrelevant records. Continue manual labelling until both classes are represented.

Agreement is not ready

Agreement collection needs an owned cloud project with accepted collaborators who have saved their screening progress.

Privacy, copyright and terms

AIM Review works locally unless you choose to log in. Signed-in data is used for authentication, session management, and saving progress in Firebase Storage. AIM Review App. Copyright (C) 2026 Sergio Mena Ortega, Paris Alexandros Lalousis & Nikolaos Koutsouleris.

Interactive tutorials

Loading the step-by-step tutorials...