%201.svg){kind=logo}
Sovereign AI for Financial Intelligence: Multi-Bank Transaction Analysis and AML Detection Using Open-Source Large Language Models.
- The paper introduces "AFM Ingestion," a five-layer architecture developed to address the challenges of fragmented banking infrastructure and strict data-residency rules in Kazakhstan.
- The system utilizes self-learning format detection for multi-bank ingestion, a canonical schema for semantic enrichment, and hybrid transaction classification.
- It features eight interpretable Anti-Money Laundering (AML) typologies that are directly mapped to the Criminal Code of the Republic of Kazakhstan for judicial traceability.
- The architecture includes a natural-language analyst interface using a locally hosted Qwen2.5-Coder-14B model to convert Russian-language queries into SQL.
.jpg)
Related Papers
Municipal Tokens as Urban Policy Tools: The Case of LVGA and the MyLugano App.
This paper examines the SwissLedger blockchain, the LVGA token, and the MyLugano app as integrated components of a publicly enabled urban infrastructure in Lugano.
.jpg)
The Unbuilt Sandbox: Regulatory Pathways for Digital Asset Payments in Reform-Stage Economies
This paper examines the "Strategic Regulatory Gray Area" (SRGA) in reform-stage economies, where regulators neither formally authorise nor prohibit digital-asset payments, and proposes a framework to transition these activities into formal regulation.