Banking Done Right: Redefining Retail Banking with Language-Centric AI

TL;DR

The paper tackles the inefficiency and risk of core banking workflows by introducing Ryt AI, an in-house LLM-powered, agentic framework that enables customers to execute core financial transactions through natural language in a regulator-approved setting. It advances a modular four-agent architecture (Guardrails, Intent, Payment, FAQ) built on the domain-specific ILMU model, with OCR capabilities, structured messaging, and a strict human-in-the-loop to ensure safety and compliance. The authors demonstrate production-grade deployment within a Malaysian digital bank, highlighting scale (tens of thousands of users and tens of thousands of transactions per month), multilingual capability, and low hallucination rates, while maintaining auditable and stateless memory design. This work provides a practical blueprint for safe, regulator-aligned AI-native banking interfaces, bridging research and real-world, high-stakes financial applications.

Abstract

This paper presents Ryt AI, an LLM-native agentic framework that powers Ryt Bank to enable customers to execute core financial transactions through natural language conversation. This represents the first global regulator-approved deployment worldwide where conversational AI functions as the primary banking interface, in contrast to prior assistants that have been limited to advisory or support roles. Built entirely in-house, Ryt AI is powered by ILMU, a closed-source LLM developed internally, and replaces rigid multi-screen workflows with a single dialogue orchestrated by four LLM-powered agents (Guardrails, Intent, Payment, and FAQ). Each agent attaches a task-specific LoRA adapter to ILMU, which is hosted within the bank's infrastructure to ensure consistent behavior with minimal overhead. Deterministic guardrails, human-in-the-loop confirmation, and a stateless audit architecture provide defense-in-depth for security and compliance. The result is Banking Done Right: demonstrating that regulator-approved natural-language interfaces can reliably support core financial operations under strict governance.

Figures (5)

• Figure 1: Banking as dialogue: Ryt AI executes financial operations through natural language.
• Figure 2: Comparison of legacy fund transfer workflows and Ryt Bank’s conversational approach. (Left) The legacy flow spans multiple screens, requires manual input and rigid navigation, and lacks semantic understanding, typically taking 30–45 seconds per transaction. (Right) Ryt AI streamlines banking by replacing multi-screen workflows with a conversational interface powered by LLM agents.
• Figure 3: Ryt AI framework. A modular, multi-agent architecture that enables coordinated collaboration among specialized agents to handle distinct banking tasks.
• Figure 4: Screenshots from Ryt AI interface. (a) Malicious inputs are blocked by the Guardrails agent to prevent unsafe interactions that could compromise system operations. (b) User intents are accurately identified and routed to downstream agents to perform banking tasks such as payment processing. (c) The system handles multimodal inputs by leveraging integrated OCR to extract and interpret relevant information from images alongside text. (d) FAQs are answered with factually grounded and context-aware responses.
• Figure 5: Comparative evaluation of five LLMs across five key performance metrics. Performance values are normalized on a [0, 1] scale.