SciAudit AI
SciAudit AI by Ardie Barry Sailis, an AI-assisted scientific evidence auditing project for biomedical research
SciAudit AI
AI-assisted scientific evidence auditing for biomedical research. Designed to help researchers move from literature overload to structured, traceable and critically examined evidence.
A researcher-built evidence system
SciAudit AI starts from a practical problem encountered directly in biomedical research: evaluating large, fragmented literatures without losing the context, limitations and uncertainty behind the conclusions.
What is SciAudit AI?
SciAudit AI is an early-stage research technology project I am developing to improve how biomedical researchers evaluate, compare and organize scientific evidence.
Development began in 2025 as an independent research initiative, initially in response to challenges I encountered in my own biomedical research workflow.
The project takes an AI-native approach to evidence analysis. Instead of treating an AI-generated summary as the endpoint, SciAudit AI is being designed around the underlying evidence: what was reported, where it was reported, how the study was conducted, how strong the support is, which studies disagree, and what limitations remain.
What I am building
The initial system is being designed around the stages of scientific evidence evaluation rather than around generic text generation.
Literature synthesis
Organize and synthesize relevant biomedical literature while preserving the relationship between conclusions and their supporting sources.
Cross-study comparison
Compare populations, exposures, interventions, outcomes, methods and findings across studies rather than treating the literature as one undifferentiated body of evidence.
Claim auditing
Examine whether scientific statements appear adequately supported, overstated, qualified or unresolved.
Methodological critique
Surface study-design limitations, measurement constraints, confounding considerations and other sources of uncertainty.
Evidence organization
Convert fragmented literature findings into structured, inspectable evidence records that researchers can revisit.
AI-assisted research
Use AI as the central computational layer for evidence analysis, workflow design and research assistance while keeping primary scientific sources visible.
What I want SciAudit AI to do differently
The distinction is not that scientific AI is new. The aim is to combine a biomedical research workflow with explicit evidence structure, provenance, disagreement and uncertainty.
Not just an AI summary.
An evidence audit.
SciAudit AI is being developed from the perspective of a biomedical researcher who repeatedly has to determine not only what papers say, but how well the evidence supports what is being concluded.
From question to auditable evidence
A visual representation of the conceptual workflow currently guiding development.
Built on a real research workflow
The first version is being developed and tested within my own biomedical research workflow. This provides a practical environment in which to test whether an AI-assisted evidence auditing process genuinely improves literature synthesis, comparison and critical assessment.
The immediate objective is not to release a polished commercial product. It is to determine which parts of the workflow are genuinely useful, reproducible and worth developing further.
This self-validation approach is intended to create a stronger foundation before the system is opened more broadly to other researchers.
Why this matters
Biomedical researchers routinely work across large and fragmented literatures in which studies differ in population, exposure or intervention, measurement methods, endpoints, statistical approaches and evidentiary strength.
A fluent literature summary can therefore be useful while still obscuring important differences between studies. The harder problem is often not finding information, but determining what can reasonably be concluded from it.
SciAudit AI is being developed to help make those distinctions easier to identify, organize and revisit.
Toward a broader release
The timeline is intentionally long because the project is being developed through validation and iteration rather than a rapid product launch.
Open to collaboration
Conversations with researchers can directly inform the next stage of the project.
Help shape the research workflow
I am interested in conversations with biomedical researchers, scientists, methodologists and technically minded collaborators who are interested in trustworthy AI-assisted scientific evidence analysis.