CLPS Incorporation Completes AI-Assisted Anti-Money Laundering Review Project for a Major Bank, Achieving Accuracy with Fine-Tuned Large Language Model

CLPS completes a Gen AI-based AML review system for a major bank, showing small fine-tuned models can outperform larger LLMs in regulated use cases.

Rhea-AI Summary

CLPS (CLPS) has completed a Generative AI-assisted Anti-Money Laundering Governance & Decision Framework Project for a leading Chinese commercial bank within a Hong Kong regulatory Gen A.I. Sandbox.

The company fine-tuned an open-source small Large Language Model to automate AML risk ratings, achieving an accuracy rate above 90%, compared with over 40% for an off-the-shelf general LLM and just over 30% for a larger model with 400% more parameters. The solution uses synthetic data augmentation, class-weighted loss, and stratified sampling, plus a seven-dimensional scorecard, LLM-as-a-judge testing, and human-in-the-loop review. CLPS plans to reuse this standardized framework in areas such as loan approval, credit monitoring, and user behavior analysis, and as a base for proprietary financial LLM products.

Positive

• Risk-rating accuracy >90% achieved by fine-tuned small LLM in AML reviews
• Fine-tuned small LLM outperformed general LLM (>40% accuracy) and larger model (~30%)
• Standardized AI framework enables reuse in loan approval, credit monitoring, and behavior analysis
• Project completed within a regulatory Gen A.I. Sandbox, with formal client commendation
• Methodology underpins planned proprietary small financial LLM products

Negative

• None.

News Explained

The completed workflow keeps final AML decisions with client experts; its reported over 90% accuracy remains constrained by the initial data set.

CLPS reports that its AML project has concluded, with the sandbox review report submitted and client business experts retaining final decision-making authority over review conclusions.

That human-in-the-loop arrangement means the disclosed system assists with risk-rating and reporting workflows but does not replace the client’s final review authority. The company describes the roadmap as production-ready, but also says the current accuracy above 90% is limited by the initial training-data scale and that improvement depends on ingesting more high-quality real-world data.

Market Context

23.95% was the 24-hour reaction to CLPS’s Aug 20 AI banking-modernization completion, a relevant pri...

Analysis

23.95% was the 24-hour reaction to CLPS’s Aug 20 AI banking-modernization completion, a relevant prior deployment record alongside this AML project’s reported accuracy. The earlier record involved a major bank, while this announcement addressed AML review.

Key Figures

Risk-rating accuracy: Over 90% Off-the-shelf LLM accuracy: Over 40% Larger model accuracy: Just over 30% +2 more

Risk-rating accuracy

Over 90%

Fine-tuned small LLM for AML review

Off-the-shelf LLM accuracy

Over 40%

Compared with fine-tuned model

Larger model accuracy

Just over 30%

Model with 400% more parameters

Parameter difference

400% more parameters

Larger model comparison

Evaluation dimensions

Seven dimensions

Model-output quality scorecard

Previous AI Reports

1 past event · Latest: Aug 20

Same Type 1 event

1. Aug 20

   AI banking modernization

   24h Move

   +23.9%

   Completed AI modernization for a major bank, reporting 98% optimized accuracy

24h Move is the share-price change in the day after each event; other market factors may also have contributed.

Key Terms

anti-money laundering, large language models, regtech, class-weighted loss functions, +1 more

5 terms

anti-money laundering regulatory

"Generative Artificial Intelligence (AI)-Assisted Anti-Money Laundering (AML)"

Anti-money laundering are rules, checks and processes banks and other financial firms use to stop criminals from hiding or moving illegal money. Think of it like ID checks and receipts in a store that make it harder to pass off stolen goods as legitimate; for investors, strong anti-money laundering controls reduce the risk of fines, shutdowns, and reputational damage that can wipe out shareholder value.

large language models technical

"fine-tuning small general-purpose Large Language Models (LLMs)"

Large language models are advanced AI systems trained on vast amounts of text to understand and generate human-like writing, like a very fast reader and writer that learns patterns in words and sentences. They matter to investors because they can change how companies operate—automating customer service, speeding analysis, cutting costs, creating new products—and they introduce risks around accuracy, security and regulation that can affect a firm’s revenue and reputation.

regtech regulatory

"development of regulatory technology (RegTech) in the banking industry"

Regtech is software and digital tools that help companies follow laws and reporting rules automatically, by monitoring transactions, flagging risks, and generating required reports. For investors, regtech matters because it reduces the chance of costly fines or business disruptions, lowers compliance costs, and can speed operations—think of it as a smart navigation and alert system that keeps a company on the legal road so capital and management focus on growth rather than paperwork.

class-weighted loss functions technical

"introduced class-weighted loss functions and stratified sampling techniques"

Class-weighted loss functions are a machine-learning training technique that assigns different penalty weights to errors on different outcome classes, so the model pays more attention to some categories (often rare or important ones). For investors, this matters because models used in trading, risk scoring or fraud detection that weight classes can reduce costly mistakes on infrequent but high-impact events by effectively telling the model to treat some errors as more serious, like prioritizing hitting a small, high-value target.

human-in-the-loop technical

"Seven-Dimensional Scorecard and Human-in-the-Loop (HITL) Safeguard"

Human-in-the-loop describes systems where people supervise, check, or make final decisions on work performed by automated tools or algorithms. Like a pilot overseeing an autopilot, humans step in to catch errors, interpret nuance, and apply judgment that machines may miss. For investors, this matters because human oversight can reduce operational and regulatory risk, improve decision quality, and increase trust in results produced by automated systems.

AI-generated analysis. How Rhea-AI works. Not financial advice.

See more from StockTitan in Google Search and AI answers. Adds StockTitan as a preferred source · opens Google

Add on Google Not now

HONG KONG, Sept. 22, 2026 /PRNewswire/ -- CLPS Incorporation (the "Company" or "CLPS") (Nasdaq: CLPS) today announced the successful completion of its Generative Artificial Intelligence (AI)-Assisted Anti-Money Laundering (AML) Governance & Decision Framework Project, implemented for a leading Chinese commercial bank (the "Client"), within a Hong Kong regulatory department's Gen A.I. Sandbox.

In response to the compliance challenges driven by continuously rising transaction volumes, the project innovatively explored the deep application of generative AI in AML review workflows. By fine-tuning small general-purpose Large Language Models (LLMs), CLPS achieved the intelligent and automated generation of risk ratings, establishing a new benchmark for the development of regulatory technology (RegTech) in the banking industry.

Significant Technical Milestones: Fine-Tuned Small General-Purpose Language Models Outperform Larger LLMs

In recent years, the complexity of transactions has surged, leaving AML review processes hindered by time-consuming manual workflows and highly subjective evaluation standards. Concurrently, the regulatory department has instituted stringent high requirements regarding the transparency, explainability, and evidence traceability of AI-assisted decision-making. Compounding these challenges, the Client's production environment operates with limited graphics processing unit (GPU) resources, making the on-premises deployment of massive LLMs structurally unfeasible.

To resolve this, the project team utilized an open-source small LLM as its foundational model. Through precise domain fine-tuning and rigorous data engineering, the team achieved an impressive risk-rating accuracy rate exceeding 90%. This result not only far surpassed the off-the-shelf, general-purpose LLM's accuracy of over 40%, but also significantly outperformed a larger model with 400% more parameters, which achieved just over 30% accuracy. This breakthrough provides the Client with a powerful, intelligent assistance tool for transaction compliance while validating a new fintech implementation paradigm: replacing massive parameters and high compute costs with compact models and robust fine-tuning.

Data Engineering Breakthroughs: Synthetic Data Augmentation

The original dataset comprised a small set of authentic, anonymized cases — a volume insufficient for effective model fine-tuning. To overcome this, the CLPS team adopted two critical strategies:

1. Risk Distribution Matching: Seed data was strictly curated to mirror the exact distribution of the Client's real-world business scenarios—a vast majority of low-risk, a moderate portion of medium-risk, and a small minority of high-risk—ensuring the model's output would not deviate from actual operational realities.
   
   
2. AI-Powered Data Augmentation: Utilizing generative rewriting strategies, the seed records were significantly expanded into a substantial collection of distinct datasets. This approach preserved core risk patterns while varying the syntactical structure, perfectly balancing data validity with semantic diversity.

To address the class imbalance inherent in datasets dominated by low-risk cases, the team introduced class-weighted loss functions and stratified sampling techniques. These adjustments significantly enhanced the model's sensitivity and precision in identifying high-risk cases.

Overcoming Generative Output Challenges: Task Decoupling and Mandatory Evidence Citation

To mitigate the common generative AI issues of "hallucinations" and context interference when processing complex workflows, the CLPS team implemented two key innovations at the model-tuning level:

1. Task Decoupling Design: The generation of risk rating, Request for Information (RFI) inquiries, and final reports were decoupled into independent output streams. This architectural shift eliminated the accuracy degradation typically caused by cross-task interference.
   
   
2. Reasoning Chain Optimization: Initial attempts to inject step-by-step reasoning rationale to improve explainability inadvertently reduced overall accuracy. The team decisively pivoted to a staged decoupling approach with class weighting. Explainability ultimately secured through a comprehensive seven-dimensional scorecard, ensuring transparency without compromising model performance.

Seven-Dimensional Scorecard and Human-in-the-Loop (HITL) Safeguard

CLPS, in collaboration with experts from the Client's legal and compliance department, developed a weighted evaluation scorecard. This framework systematically measures model output quality across seven dimensions: completeness of case information, reasonableness of risk assessment, depth and logic of analysis, adequacy of suspicious risk assessment, degree of report structuring, thoroughness of investigation description, and clarity of results.

The evaluation system also integrated a third-party LLM to conduct blinded A/B testing (LLM-as-a-Judge). By randomizing the output order of fine-tuned versus non-fine-tuned models during the scoring process, the team effectively eliminated positional bias.

For extreme edge cases, the project strictly adhered to HITL mechanism. The Client's business experts retain final decision-making authority, ensuring every review conclusion remains entirely traceable and auditable, fully satisfying the regulatory department's rigorous regulatory requirements for evidence traceability.

Client Feedback and Strategic Outlook

The Client's business units have highly praised the model's efficacy, explainability, and deployment feasibility. By requiring only the deployment of a small LLM under strict resource constraints, the project drastically lowers the barrier to entry for actual production deployment. In a formal letter of commendation, the Client emphasized that since the project's launch this March, the CLPS team collaborated seamlessly with the Legal and Compliance Department and the Technology Department of the Client's Hong Kong branch. Together, the teams navigated complex technical challenges to execute case screening, data cleaning, model training, fine-tuning, inference, and testing, culminating in the successful submission of the Gen A.I. Sandbox review report to the regulatory authority.

Throughout the initiative, the CLPS team demonstrated exceptional professionalism, deep AI and AML domain expertise, and a rigorous commitment to project delivery. Notably, the team achieved results that significantly exceeded expectations in model optimization and technical problem-solving. By elevating the accuracy of the suspicious transaction case analysis model to over 90% using a fine-tuned small LLM, CLPS fully demonstrated its robust capabilities in the fine-tuning and deployment of specialized financial AI models. Following a comprehensive knowledge transfer session, this collaboration project concluded successfully. The Client expressed its sincere gratitude to the CLPS project team for their dedicated effort and professional support, noting that it looks forward to deepening its partnership with CLPS across future innovative fintech and AI initiatives.

CLPS recognizes that the current accuracy rate of over 90% is inherently limited by the scale of the initial training data; future performance improvements will be driven by the continuous ingestion of high-quality, real-world business data.

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 across the Greater Bay Area and globally as they navigate cross-border compliance complexities.

Furthermore, CLPS has distilled this project into a standardized financial AI model training framework. This highly replicable methodology can be scaled horizontally across other banking business scenarios, including:

• Loan Approval Assistance: Automated risk assessment and credit recommendation generation based on the multidimensional data of corporate and individual borrowers.
  
  
• Credit Monitoring: Continuous risk monitoring and early-warning detection for existing credit asset portfolios.
  
  
• User Behavior Analysis: Identification of anomalous transaction patterns and potential compliance risks.

This scalable methodology will also serve as the foundation for CLPS's upcoming R&D initiatives aimed at launching small proprietary financial LLM products, reinforcing the Company's commitment to providing low-cost, highly available RegTech solutions to the global banking sector.

Mr. Raymond Lin, Chief Executive Officer of CLPS, said: "This deep collaboration with the Client represents a pivotal milestone for CLPS in the financial AI sector. The project's success not only validates the commercial viability of compact models with robust fine-tuning in vertical use cases, but also highlights a fundamental industry truth: 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. We are deeply grateful for the Client's trust and partnership, which made this meaningful technical validation within the regulatory department's Gen A.I. Sandbox possible. Moving forward, CLPS will leverage this success to accelerate our investments in financial AI R&D, driving our strategic transition from project-based delivery to product-driven enablement, and equipping more financial institutions with pragmatic, efficient, and trustworthy AI solutions."

About CLPS Incorporation

CLPS Incorporation (NASDAQ: CLPS), established in 2005 and headquartered in Hong Kong, is at the forefront of driving digital transformation and optimizing operational efficiency across industries through innovations in artificial intelligence, cloud computing, and big data. Our diverse business lines span sectors including fintech, payment and credit services, e-commerce, education and study abroad programs, and global tourism integrated with transportation services. Operating across 10 countries worldwide, with strategic regional hubs in Shanghai (mainland China), Singapore (Southeast Asia), and California (North America), and supported by subsidiaries in Japan and the UAE, we provide a robust global service network that empowers legacy industries to evolve into data-driven, intelligent ecosystems. For further information regarding the Company, please visit: https://ir.clpsglobal.com/, or follow CLPS on Facebook, Instagram, LinkedIn, X, and YouTube.

Forward-Looking Statements

Certain of the statements made in this press release are "forward-looking statements" within the meaning and protections of Section 27A of the Securities Act of 1933, as amended, and Section 21E of the Securities Exchange Act of 1934, as amended. Forward-looking statements include statements with respect to the Company's beliefs, plans, objectives, goals, expectations, anticipations, assumptions, estimates, intentions, and future performance. Known and unknown risks, uncertainties and other factors, which may be beyond the Company's control, may cause the actual results and performance of the Company to be materially different from such forward-looking statements. All such statements attributable to us are expressly qualified in their entirety by this cautionary notice, including, without limitation, those risks and uncertainties related to the Company's expectations of the Company's future growth, deployment in the AI technology sector, performance and results of operations, the Company's ability to capitalize on various commercial, M&A, technology and other related opportunities and initiatives, as well as the risks and uncertainties described in the Company's most recently filed SEC reports and filings. Such reports are available upon request from the Company, or from the Securities and Exchange Commission, including through the SEC's Internet website at http://www.sec.gov. We have no obligation and do not undertake to update, revise or correct any of the forward-looking statements after the date hereof, or after the respective dates on which any such statements otherwise are made.

Contact:

CLPS Incorporation
Rhon Galicha
Investor Relations Office
Phone: +86-182-2192-5378
Email: ir@clpsglobal.com

View original content:https://www.prnewswire.com/news-releases/clps-incorporation-completes-ai-assisted-anti-money-laundering-review-project-for-a-major-bank-achieving-accuracy-with-fine-tuned-large-language-model-302886033.html

SOURCE CLPS

FAQ

AI-generated questions and answers. How Rhea-AI works. Not financial advice.

How did CLPS improve model performance with a relatively small initial dataset?

CLPS started from a small set of anonymized real cases and expanded it using risk distribution matching and AI-powered data augmentation. Seed data was curated to mirror actual business risk proportions, then generative rewriting created many distinct records that preserved core risk patterns while varying syntax. Class-weighted loss functions and stratified sampling were applied to handle class imbalance and improve sensitivity to high-risk cases.

What measures were taken to reduce hallucinations and context interference in the AI outputs?

The project adopted a task decoupling design, separating risk rating, Request for Information generation, and final report drafting into independent output streams to avoid cross-task interference. When step-by-step reasoning initially reduced accuracy, the team shifted to staged decoupling with class weighting and used a seven-dimensional scorecard for explainability instead, helping to control hallucinations while maintaining performance.

How is the quality of the AML model’s output evaluated and validated?

Model output quality is measured with a weighted seven-dimensional scorecard covering completeness, risk assessment reasonableness, depth and logic of analysis, adequacy of suspicious risk assessment, report structuring, investigation thoroughness, and clarity of results. A third-party LLM performs blinded A/B testing (LLM-as-a-judge) by randomizing output order from fine-tuned versus non-fine-tuned models to remove positional bias.

What role do human experts play in the AML review process after AI deployment?

For extreme edge cases, the project applies a human-in-the-loop (HITL) mechanism. Business experts from the client retain final decision authority, ensuring each review conclusion is traceable and auditable and meets the regulatory department’s strict requirements for evidence traceability.

How does CLPS plan to use this AML project for future business and product development?

CLPS has turned the AML initiative into a standardized financial AI model training framework that can be scaled to other banking scenarios such as loan approval assistance, credit monitoring, and user behavior analysis. This framework will also support upcoming R&D aimed at launching small proprietary financial LLM products to deliver low-cost, highly available RegTech solutions.

What limitations or next steps did CLPS identify for the AML model’s performance?

CLPS noted that the current accuracy of over 90% is constrained by the scale of the initial training data. The company expects future improvements to come from the continuous ingestion of high-quality, real-world business data, which should further enhance model performance over time.