Several tools in Nested Knowledge use artificial intelligence and machine learning to make systematic reviews faster and more effective to conduct. This page discloses, for each tool: what it does, whether it is optional, which of your data it uses, how the underlying model works, and whether your data is used to train it.
Every AI tool in Nested Knowledge is designed around the same principle: AI output is presented for human review, and nest owners retain final responsibility for all decisions in their review. For guidance on responsible use, see Guidance on using AI in Nested Knowledge and our RAISE Compliance Guide.
1. Smart Search #
Smart Search prompts user input of a research question, encourages refinement of that question, and generates varying-sized corresponding search strings for use in searching PubMed.
Optional: Smart Search is an alternative to writing a Boolean search manually; it is never applied without user initiation.
a. Which data does Smart Search use? #
Smart Search uses only the text entered as the research question and, where applicable, responses to prompts used to refine the research question.
b. How does the model work? #
Smart Search is an agent-based LLM system that employs an iterative generator-critic loop to generate and assess Boolean strings based on user-entered textual research questions and iterative chat-based user clarifications. See our full Smart Search model card, which also uses Research Question Refinement.
This model is not trained or updated from your data.
2. RoboPICO #
RoboPICO is used to:
- Highlight Populations, Interventions/Comparators, and Outcomes in abstracts during Screening.
- Optional: toggled on by default, but can be toggled off and will remain off for all abstracts in the queue/ongoing modules and Study Inspector until toggled back on.
- Generate the most commonly reported terms among the literature, to inform the building of search queries in Search Exploration.
- Optional: Search Exploration is not a required step in AutoLit. However, when it is used, RoboPICO auto-generates terms when concepts are entered and “Refresh Exploration” is selected; this cannot be switched off within Search Exploration.
a. Which data does RoboPICO use? #
The abstract of records is made available to the model.
b. How does the model work? #
RoboPICO uses a fork of the machine learning system offered by RobotReviewer. Specifically, Named-entity Recognition models extract Patient/Problem, Intervention, and Outcome entities from data in article abstracts. NK’s modifications to RobotReviewer are open and General Public Licensed.
This model is not trained or updated from your data.
3. Bibliomine #
Bibliomine extracts references from full-text PDFs. Typically, previous systematic reviews or landmark studies are bibliomined, importing all cited references as records directly into your nest. This feature does not access your data unless you use it.
Optional: This feature is helpful if, for example, you are performing an update on an existing review, but is not required to successfully upload records to a nest.
a. Which data does Bibliomine use? #
Bibliomine consumes any PDFs uploaded for its purposes. It writes records with full bibliographic data to your nest (pending user addition).
b. How does the model work? #
Bibliomine uses Cermine, an open-source machine learning library for mining documents. Using the DOI and title extracted, full bibliographic data is retrieved from PubMed or CrossRef (in that order of preference).
This model is not trained or updated from your data.
4. Robot Screener #
Robot Screener uses a model trained on human screening decisions to make reviewer-level screening decisions based on inclusion probabilities. In effect, it replaces one human reviewer when turned on in nests with a Dual Screening mode. This feature does not have access to your data unless you turn it on in Settings.
Optional: This feature is helpful to speed up the Screening process, but is not required to successfully Dual Screen all records in a nest.
a. Which data does Robot Screener use? #
The following data from your records are model inputs:
- Bibliographic data
- Time since publication of the record
- Page count
- Keywords/Descriptors
- Abstract content
- N-grams
- OpenAI text embedding (ada-002)
- Citation counts from scite, accessed using the DOI
- Number of citing publications
- Number of supporting citation statements
- Number of contrasting citation statements
The model is trained on adjudicated abstract or full-text screening decisions, depending on your screening mode, including the exclusion reason. The model outputs a screening decision for each record requiring a reviewer-level decision.
If enabled, Robot Screener will continuously screen new records as they are imported into your nest.
b. How does this model work? #
Learn more about the screening model that generates inclusion/advancement probabilities for Robot Screener. A probability threshold, optimized on a geometric mean of precision and recall in a cross validation, determines whether records should be included or excluded.
This model is only available within your nest, and its parameters are not shared with other nests or users.
5. Smart Screener (Criteria-Based Screening) #
Criteria-Based Screening (CBS) allows users to input their inclusion criteria as yes/no questions; Smart Screener answers these questions using the study abstract and/or full text, attaching an annotation. Users are encouraged to review these answers before manually assigning a screening decision, but may bypass this by applying a threshold of criteria met in order to include/exclude studies in bulk.
Optional: Smart Screener only runs when initiated by a user with Criteria-Based Screening configured. Users control which records it runs on as the run can be scoped to any filtered set of records. Any generated answer can be edited or cleared before a screening decision is made.
a. Which data does Smart Screener use? #
Answers are generated for abstracts or full-text PDFs at the user’s discretion. You instruct the model through:
- Criteria Name
- Question Contents
b. How does the model work? #
Smart Screener uses an OpenAI LLM to identify relevant concepts or answer questions about plain-text reductions of abstracts and full-text PDFs. The model’s recommendations are related to the abstract/PDF as an annotation using fuzzy (edit distance) search.
The model, and OpenAI in general, is not trained or updated using your data.
6. Core Smart Tags #
Core Smart Tags (CSTs) prompts user input of a research question, encourages refinement of that question, and builds a tag hierarchy/data extraction template of PICOs, Study Type, Study Location, and Study Size based on all study abstracts. It also extracts this data for review.
Optional: Core Smart Tags only runs when initiated by a user.
a. Which data does Core Smart Tags use? #
Core Smart Tags uses each study’s title, abstract, and bibliographic data to determine the appropriate categorization and extraction of the given concepts.
b. How does this model work? #
Core Smart Tags is an integrated system for extracting, analyzing, and structuring information from biomedical research abstracts. It combines independent models and tools to generate a comprehensive overview of study characteristics:
- Hierarchy Tool: Generates a hierarchical structure of entities based on PICO extractions to allow users to navigate studies by specific attributes (e.g., interventions > vascular surgeries > thrombectomies).
- Smart Study Size (SSS): Extracts the sample size from abstracts with a heuristic-based algorithm.
- Smart Study Location (SSL): Identifies the geographic location of studies by extracting references to countries and cities.
- Smart Study Type (SST): Classifies abstracts into predefined study types (e.g., clinical, cohort, observational).
- BioELECTRA-PICO: Extracts PICO (Participants, Interventions, Outcomes) elements using a fine-tuned transformer model.
Additionally, Research Question Refinement is used.
This model is not trained or updated from your data.
7. Adaptive Smart Tags #
Adaptive Smart Tags (ASTs) are context-aware recommendations of relevant concepts in study abstracts and/or full texts using an OpenAI Large Language Model (LLM). These recommendations include provenance via an annotation within the abstract or PDF, making review of the model’s work possible. The option is also provided to bypass this review and auto-extract relevant data for rapid exports.
Optional: This feature is helpful to speed up the data extraction process, but is not required to perform data extraction of all records in the nest. Users control which records it runs on as the run can be scoped to any filtered set of records. Any generated answer can be edited or cleared from a study. We also offer an optional high-fidelity mode that has LLM-assisted PDF-to-plain-text conversion for extraction of tables and figures in addition to text.
a. Which data does Adaptive Smart Tags use? #
Recommendations are generated for abstracts or full-text PDFs at the user’s discretion. You instruct the model through:
- Tag Names
- Hierarchical structure of tags
- Question Types
- Question
- Contents (this includes text, table, options, and numerical contents created alongside tags)
b. How does the model work? #
Adaptive Smart Tags uses an OpenAI LLM to identify relevant concepts or answer questions about plain-text reductions of abstracts and full-text PDFs. The model’s recommendations are related to the PDF as an annotation using fuzzy (edit distance) search.
The model, and OpenAI in general, is not trained or updated using your data.
8. Smart Critical Appraisal #
Smart Critical Appraisal (SCA) reads the full text of included studies and recommends answers to the questions in your configured critical appraisal system(s), annotating the passage(s) supporting each recommendation. Every recommendation is presented for reviewer acceptance, revision, or clearing before it counts, and can be traced back to the source text.
Optional: SCA only runs when initiated by a user from the Critical Appraisal module, and never overwrites existing responses (human or previously generated).
a. Which data does Smart Critical Appraisal use? #
SCA uses:
- The full-text PDFs of included studies (studies without an uploaded full text are skipped; abstracts alone are not used)
- The questions of the critical appraisal system(s) configured in Nest Settings
- Each study’s design, to automatically match the study to the appropriate configured appraisal system
b. How does the model work? #
SCA uses an OpenAI LLM that evaluates each checklist question against the extracted full text of the study, one question and one document at a time, and returns a structured response with passage-level annotations. See the Smart Critical Appraisal model card for intended use, validation metrics, and limitations.
Full texts are processed transiently for appraisal generation and are not retained beyond the workflow. Your documents are not used to train the model.
9. Smart Insights #
Smart Insights generates written, evidence-backed insights from your extracted data. Each generated insight is composed of a summary (Insight), the individual Claims within it, and the study citations and highlighted evidence (Evidence) supporting each claim, so every sentence remains traceable to its source material.
Optional: Smart Insights only runs when initiated by a user from the Export page or Dashboard Editor. Users control which tags insights are generated on as generation can be scoped to any selection of tags. Any generated insight can be edited or deleted before use.
a. Which data does Smart Insights use? #
Smart Insights uses:
- Your tag hierarchy (Single Apply question tags only)
- The tagged contents extracted against those tags, including data tables
- The bibliographic data of the studies underlying each claim, used for citation
b. How does the model work? #
Smart Insights uses an OpenAI LLM and works only with evidence in your project — it does not draw on outside sources — and maintains a link between each insight and its underlying source data.
This model is not trained or updated from your data.
10. Smart Abstract #
Smart Abstract generates a structured abstract of your nest — Methods, Results, and Discussion — drawn entirely from the data in your review. The Methods section is templated rather than AI-generated, so key methodological details (including every AI tool used in the review) are reported every time an abstract is generated. Available to Enterprise and Academic+ subscribers.
Optional: Smart Abstract only runs when initiated by a user from the Abstract Editor, and warns before overwriting an existing abstract.
a. Which data does Smart Abstract use? #
Smart Abstract uses:
- Your Research Question
- Your nest’s workflow configuration (searches, screening and tagging configuration, and any Smart features used), which populates the templated Methods section
- Search, screening, and critical appraisal outcomes
- Generated Insights and Smart Insights, where available
- A summary of extracted data, where Insights have not been generated
Because Smart Abstract draws on the outputs of other tools in your review, including screening decisions, AI-generated tags and Insights, the user controls described in those sections carry through. Any content that has been edited, cleared, or excluded is reflected accordingly in the generated abstract, and the abstract itself remains fully editable.
b. How does the model work? #
Smart Abstract uses an OpenAI LLM. The Methods section is produced by templating from your nest configuration, the Results and Discussion sections are generated by a large language model grounded exclusively in your nest’s data; the system does not draw on outside sources and deliberately avoids editorializing or padding.
This model is not trained or updated from your data.
Synthesis AI Disclosure #
The records included in Synthesis, as well as the tags applied to underlying reports, may have been collected by users with the assistance of artificial intelligence. AI tools in Nested Knowledge are integrated with direct researcher oversight, and nest owners have final responsibility for the accuracy of all Screening, Tagging, Extraction, and Critical Appraisal.
Data Security & Protection using AI #
At Nested Knowledge, we understand the importance of safeguarding your data. We have robust security measures in place designed to keep your information private, secure, and accessible. Nested Knowledge internally defines and adheres to numerous security policies designed to maintain the security and privacy of user data while maintaining high uptime with strong disaster recovery measures.
In regard to our database, data is gated behind standard, best-practice authentication and authorization measures, and each project is logically separated from other projects. All network communications, within our app and to the client, are encrypted. Our sub-processors (e.g. OpenAI) are held to a data processing agreement (DPA) and reviewed prior to addition to our stack. We do not work with sub-processors who repurpose or resell your data.
Please contact us if you require further information.
