RegulaGraph

Evidence-Grounded Regulatory Investigation with Meta Muse Regulatory documents rarely exist in isolation. A new circular may reference, amend, or repeal earlier provisions, and understanding those relationships often requires manually tracing multiple documents. RegulaGraph is an AI-assisted investigation tool designed to make that process more transparent. Built around public Central Bank of Tunisia regulatory documents, RegulaGraph uses Meta Muse to propose typed relationships between provisions. Rather than treating generated answers as verified facts, the system checks whether supporting quotations can be traced to the source documents through exact text spans, page references, and source integrity checks. The investigation workflow separates three responsibilities: - AI proposal: Meta Muse generates candidate regulatory relationships and supporting evidence. - Deterministic verification: The application checks evidence provenance and structural consistency without claiming to establish semantic or legal correctness. - Human review: Proposed relationships remain candidates until a reviewer assesses them. The project includes an interactive investigation interface, evidence inspection, adversarial evaluation cases, and a replay-only public demonstration. The recorded video also demonstrates live Meta Model API inference. Evaluation exposed an important limitation: a model can propose an incorrect relationship while citing genuine source text. RegulaGraph makes that distinction visible instead of presenting every well-cited answer as trustworthy. Our goal is to make AI-assisted regulatory investigation more traceable, inspectable, and accountable—not to replace legal expertise or human judgment.