Article 11 Technical Documentation

The CreditIQ Engine (v3.2, May 2026) is a high-risk artificial intelligence system developed and placed on the market by NovaMind Analytics Ltd (hereinafter "the Provider") for the purpose of automated creditworthiness assessment of natural persons applying for consumer credit through the Provider's digital lending platform. The System falls within the scope of Regulation (EU) 2024/1689 (the "AI Act") and is classified as high-risk pursuant to Annex III, Section 5(b) thereof, which designates AI systems used to evaluate the creditworthiness of natural persons or to establish their credit score as high-risk, with the exception of systems detecting financial fraud.

The intended purpose of the System is to analyse applicant financial history, transaction patterns, and submitted documentation in order to generate a quantitative credit risk score on a scale of zero to one thousand, assign applicants to one of four risk categories (Low, Medium, High, or Very High), and produce a preliminary approval or decline recommendation accompanied by a supporting rationale summary. The System operates as a decision-support tool within the Provider's consumer lending workflow, where its outputs inform but do not autonomously determine the final credit decision communicated to applicants.

The deployment environment comprises dedicated cloud infrastructure hosted within the Amazon Web Services EU-Frankfurt region (eu-central-1), ensuring that all personal data processing occurs within the territory of the European Union in compliance with Regulation (EU) 2016/679 (the "GDPR"). The System is accessed by authorised loan officers at NovaMind Lending via the Provider's internal operator interface, which communicates with the inference cluster through authenticated and encrypted API endpoints.

The Provider maintains full operational control of the System and does not license or distribute it to third-party deployers, operating as both provider and deployer within the meaning of Article 3 of the AI Act.

The CreditIQ Engine employs a hybrid machine learning architecture comprising a neural network feature extraction layer followed by a gradient-boosted decision tree ensemble (XGBoost) as the primary predictive model. The neural network component serves as a learned feature transformation layer that processes raw financial inputs — including time-series transaction data, categorical employment and income variables, and structured credit bureau records — into a dense, lower-dimensional feature representation optimised for downstream credit risk prediction.

The data flow architecture of the System proceeds through five sequential stages. In the ingestion stage, applicant data is received from the Provider's online lending platform via a RESTful API endpoint secured with TLS 1.3, validated against a strict input schema, and assigned a unique processing identifier. In the enrichment stage, the System retrieves supplementary credit bureau records from Experian and Schufa via secure API integrations. In the preprocessing stage, the combined dataset undergoes standardised transformations including numerical scaling, categorical encoding, and temporal feature engineering applied to the twelve-month transaction history. In the inference stage, the preprocessed data passes through the neural network feature extractor and subsequently through the XGBoost ensemble to produce the credit risk score, risk category, and preliminary recommendation. In the output stage, results are serialised, logged in the automated event logging subsystem per Article 12, and returned to the operator interface.

The inference infrastructure is deployed on a dedicated cluster of Amazon Web Services ml.p3.8xlarge GPU instances within eu-central-1, configured for auto-scaling to support up to two hundred concurrent credit evaluations while maintaining the Provider's target latency of under two seconds per inference request. The infrastructure operates under a 99.95 per cent uptime service level agreement.

The System does not incorporate any pre-trained general-purpose AI models or third-party foundation models. Third-party software libraries incorporated include: XGBoost (v2.1.x, Apache 2.0), PyTorch (v2.3.x, BSD), and scikit-learn (v1.5.x, BSD). All versions are pinned and subject to the Provider's software supply chain security review process.