Projects using Dust in ZagrebProjects using Dust in ZagrebScientific Validator: AI Document Integrity Review Tool
Scientific Validator is a prototype for reviewing whether a report, paper, or AI-generated document is structurally trustworthy before a human relies on it.
The system checks documents across multiple integrity layers:
Reasoning structure
Source and citation traceability
Method consistency
Core inquiry continuity
Unsupported or overconfident claims
Possible fabricated or weakly grounded references
Polished language that hides weak logic
The product challenge was to turn a complex validation process into a clear, usable interface. Instead of giving one vague “AI score,” the tool breaks the review into separate report cards so the user can see which parts of the document are sound, which parts need review, and where the reasoning may fail.
My role covered product architecture, validation logic, AI workflow design, UX structure, report framework, and demo narrative.
Key focus areas:
AI document review
Source-grounded validation
Human-in-the-loop review
Research/report integrity
Structured AI outputs
Multi-agent review workflows
Clear UX for complex reasoning checks
