CLPS Incorporation has developed a fine-tuned small AI model that achieved over 90% accuracy in anti-money laundering (AML) reviews for a leading Chinese commercial bank. This breakthrough demonstrates the potential of compact models in regulated financial use cases.
TL;DR
- CLPS's fine-tuned small AI model achieved over 90% accuracy in AML reviews, outperforming larger LLMs.
- The model was developed within a Hong Kong regulatory department's Gen AI Sandbox, addressing compliance challenges and GPU resource limitations.
- This success validates the use of compact models with robust fine-tuning for vertical use cases in the financial industry.
What happened
CLPS Incorporation completed a Generative AI-Assisted Anti-Money Laundering (AML) Governance & Decision Framework Project for a leading Chinese commercial bank. The project was implemented within a Hong Kong regulatory department's Gen AI Sandbox.
The project aimed to address compliance challenges driven by rising transaction volumes and stringent regulatory requirements. CLPS fine-tuned a small general-purpose Large Language Model (LLM) to achieve an impressive risk-rating accuracy rate exceeding 90%.
This result significantly outperformed off-the-shelf general-purpose LLMs, which had an accuracy of over 40%, and larger models with 400% more parameters, which achieved just over 30% accuracy.
Why it matters
This breakthrough provides a powerful, intelligent assistance tool for transaction compliance, validating a new fintech implementation paradigm: replacing massive parameters and high compute costs with compact models and robust fine-tuning.
The success of this initiative solidifies CLPS's technological leadership in financial RegTech and provides a pragmatic, scalable, and production-ready AI roadmap for financial institutions globally.
The project demonstrates that financial institutions do not necessarily need general-purpose LLMs with massive parameters. Instead, they require specialized AI capabilities tailored to precisely solve business pain points while keeping computing costs manageable.
Key facts
- CLPS's fine-tuned small AI model achieved over 90% accuracy in AML reviews.
- The model outperformed off-the-shelf general-purpose LLMs with an accuracy of over 40%.
- The model also outperformed larger models with 400% more parameters, which achieved just over 30% accuracy.
- The project was implemented within a Hong Kong regulatory department's Gen AI Sandbox.
- CLPS used data engineering breakthroughs, including synthetic data augmentation and class-weighted loss functions, to enhance model performance.
- The project integrated a seven-dimensional scorecard and Human-in-the-Loop (HITL) safeguard to ensure transparency and compliance.
- The Client praised the model's efficacy, explainability, and deployment feasibility, noting a seamless collaboration with CLPS.
- CLPS plans to scale this methodology horizontally across other banking business scenarios and launch small proprietary financial LLM products.
Context
The financial industry faces increasing compliance challenges driven by rising transaction volumes and stringent regulatory requirements. Traditional AML review processes are hindered by time-consuming manual workflows and highly subjective evaluation standards.
The success of CLPS's fine-tuned small AI model demonstrates the potential of compact models in addressing these challenges. It validates a new fintech implementation paradigm that focuses on specialized AI capabilities tailored to specific business pain points.
This breakthrough has significant implications for the financial industry, providing a pragmatic, scalable, and production-ready AI roadmap for institutions globally. It also highlights the importance of robust fine-tuning and data engineering in developing effective AI solutions for regulated use cases.