How Barie Conducts a Systematic Literature Review on GLP-1 Agonists for Weight Management — Last 3 Years of Evidence, Live Databases, Every Paper Sourced

The problem with asking any AI tool about medical evidence

A clinical researcher asks a general-purpose LLM to summarise the evidence base for GLP-1 agonists in weight management. The output lists five clinical trials, describes them accurately, and synthesises the conclusions. But it is not a systematic review. It is a highly sophisticated literature search. It is working with information retrieved from the open web.

The researcher does not know if these are the *best* five trials, the *only* five trials, or just the five trials most heavily cited on the internet. She does not know what search terms the model used, which databases it queried, or what inclusion criteria it applied. The output is a black box. If she submits a literature review based on that output to a peer-reviewed journal, it will be desk-rejected. The PRISMA flow diagram requires transparency.

Medical research demands an auditable trail of evidence. If a study is excluded, the researcher needs to know why. If a claim is made, the researcher needs to trace it back to the exact paragraph in the primary literature. A language model chatting with a search engine cannot do this. A dedicated systematic review workflow executing API queries directly against medical databases can.

Your prompt

One sentence. Barie activates the database connectors, executes boolean queries across all of them, deduplicates the results, and applies PRISMA screening methodology. Here is the full execution chain.

Step 1: Deep Research searches seven live medical databases simultaneously

The Deep Research agent converts the natural language prompt into structured boolean queries optimized for the syntax of each database. It queries all seven databases at the same time. A manual search across these databases sequentially — translating the terms, executing the queries, and exporting the results — typically takes 4-8 hours of highly repetitive work. Barie completes the retrieval phase in 45 seconds.

Step 2: Systematic screening — inclusion criteria applied, duplicates removed, studies classified

The combined retrieval yields 914 total records. Barie executes the PRISMA workflow automatically. It deduplicates across databases. It reads every abstract (and full text where available) against the inclusion criteria inferred from the prompt. It classifies the study type. The result is a clean, deduplicated, classified list of included studies ready for synthesis. This step typically takes a human researcher days to weeks.

Step 3: The structured review — every study extracted, classified, and source-linked

The PRISMA screening pipeline yields an included dataset of 84 studies. This list is not a bibliography. Barie extracts the key methodology, endpoints, and conclusions for every single study. The full dataset is formatted as a sortable, exportable matrix. Below is a sample of the highly-cited pivotal RCTs and meta-analyses that form the backbone of the evidence base in the final synthesis.

Step 4: Key findings synthesized from the evidence base — strength of evidence classified

The final review reads all 84 included abstracts and full texts and synthesises the findings into core themes. Unlike an LLM chatting with the internet, Barie classifies the evidence strength behind each finding based on the study type and sample size underlying the claim. A finding supported by three large RCTs is flagged as Strong Evidence. A finding supported by small observational data is flagged as Weak Evidence. You do not just get a summary; you get a critical appraisal.

Step 5: The review delivered in formats your research team actually uses

The full systematic review is useless as text in a chat window. It needs to be integrated into your drafting workflow. The structured review lands in Notion as a database for ongoing team collaboration. The full text, PRISMA flow data, and synthesis narrative export directly to a formatted Google Doc or Word document ready for editing. The complete matrix of 84 included studies exports to Google Sheets for sub-analysis or inclusion in systematic review appendices. The key findings and search strategy push to Slack or Teams for immediate team review.

For researcher teams using reference management software, the included list of 84 studies is exported to standard format files compatible with Zotero, Mendeley, and EndNote. Barie pulls the exact metadata directly from the database APIs (PubMed, Embase, etc.) to ensure the citations are clean, complete, and perfectly formatted. You do not spend three hours downloading RIS files from different databases and merging them manually. The clean citation file is attached to the output.

What you get

A systematic review is not just a summary of medical studies. It is a formal, reproducible method of identifying all available evidence for a specific clinical question, filtering it transparently, and synthesizing the findings based on methodological strength. A general-purpose LLM chatting with the internet cannot perform this task. It guesses. It hallucinates citations. It cannot reproduce its own work. It lacks the Boolean query syntax needed to retrieve comprehensively, the strict inclusion screening logic needed to filter objectively, and the study design classification capability needed to assess quality.

Barie’s Deep Research mode is built for researchers. It queries the actual medical databases. It retrieves the actual DOIs. It executes the PRISMA workflow step-by-step. Every claim is sourced. Every exclusion is logged. Every finding is graded by evidence quality. It turns a process that takes a human researcher three weeks into a process that takes forty-five minutes.

Barie features used in this task