FinRAG-12B: A Production-Validated Recipe for Grounded Question Answering in Banking

TL;DR

This paper introduces FinRAG-12B, a grounded, domain-specific language model for banking that achieves high accuracy, reliable citations, and safe refusal, optimized for real-world deployment with significant performance and cost benefits.

Contribution

The paper presents a novel data-efficient training pipeline, a calibrated refusal mechanism, and an end-to-end deployment methodology for grounded banking LLMs, outperforming GPT-4.1 in key metrics.

Findings

FinRAG-12B outperforms GPT-4.1 in citation grounding accuracy.

The calibrated refusal mechanism reduces unsafe responses to 12%.

Deployment at financial institutions improves query resolution by 7.1 percentage points.

Abstract

Large language models (LLMs) are rapidly being adopted across various domains. However, their adoption in banking industry faces resistance due to demands for high accuracy, regulatory compliance, and the need for verifiable and grounded responses. We present a unified, data-efficient framework for training grounded domain-specific LLMs that optimizes answer quality, citation grounding, and calibrated refusal under real-world deployment constraints. First, we describe a data generation pipeline that combines LLM-as-a-Judge filtering, citation annotation, and curriculum learning with only 143M tokens. The resulting 12B model achieves high answer quality outperforming GPT-4.1 on citation grounding, with a modest citation tradeoff versus the untuned base. Second, we propose a calibrated refusal mechanism: training on 22% unanswerable examples yield a 12% "I don't know" rate, substantially improving over the base model's unsafe 4.3% rate while avoiding GPT-4.1's over-refusal (20.2%). Third, we present an end-to-end methodology spanning from data curation to quantized serving. The system is deployed at 40+ financial institutions, achieving a 7.1 percentage point improvement in query resolution (p < 0.001). Additionally, the model delivers 3-5x faster responses at 20-50x lower cost compared to GPT-4.1.