Abstract

Regulated industries — banking, insurance, capital markets, life sciences, and supply chain — are subject to mandatory data submissions governed by frameworks such as the Basel Committee's BCBS 239 principles, the Federal Reserve's Comprehensive Capital Analysis and Review (CCAR), FinCEN anti-money-laundering (AML) obligations, HIPAA, and FDA 21 CFR Part 11. Existing tooling for validating data against these obligations falls into four inadequate categories: manual spreadsheet-based checks that do not scale, commercial data-quality suites that require multi-month implementations and dedicated engineering teams, brittle in-house scripts tied to individual engineers, and vendor-hosted SaaS platforms that raise data-residency concerns for regulated data. This paper presents the architecture of DataGuard, a self-hosted, AI-accelerated data quality and governance platform that integrates five technology domains — large language model (LLM) assisted rule authoring, an eight-database enterprise connectivity layer, cross-source column-level data lineage analysis, a library of seven regulatory rule packs spanning five industry sectors, and cloud-native, zero-infrastructure deployment — into a single deployable application. We describe the platform's layered-monolith architecture, its REST API surface, its four distinct Claude-API-based AI subsystems (including an automated metadata cataloging engine, DocuSmart), its three-tier column-matching algorithm for lineage inference, and its security and governance design, including an OAuth-based Git integration that establishes a version-controlled chain of custody for validation rules and results. We further situate the platform against alternative approaches, walk through an end-to-end validation session on the live application, and discuss the architecture's limitations and directions for future work. The platform is publicly accessible and its source is version-controlled, allowing independent verification of the claims described herein.