SquareGen — LLM-based Credit Scoring

What

SquareGen is a credit scoring product that uses fine-tuned large language models instead of classical gradient boosting models for lending decisions. It is built for lenders and credit risk teams that want to score applicants with fewer input features while maintaining or improving predictive performance and explainability.

The product appears positioned as a production-oriented scoring layer for banks, fintechs, and SME or consumer lenders operating in regulated environments. Its workflow is straightforward: a client shares labeled historical data, SquareGen trains and deploys a model, and the client consumes scores through a cloud API or a self-hosted deployment.

Features

• LLM-based credit scoring models — Uses fine-tuned LLMs for underwriting, with the company claiming better AUC than gradient boosting in some use cases while requiring fewer features.
• Lower-feature scoring — Reduces the number of required inputs, which can lower applicant friction and reduce bureau or API data acquisition costs.
• Built-in explainability outputs — Returns top features, attention-based signals, drop-one-out analysis, and an AI-generated risk rationale to support score interpretation.
• API and deployment flexibility — Offers managed cloud scoring via API and Python SDK, with self-hosted, on-premise, and air-gapped deployment options for tighter data control.
• Client-specific model training — Fine-tunes models per customer rather than on a shared model, which the company presents as a way to limit cross-client data leakage.
• Fast proof-of-concept process — States that a PoC can be delivered within 72 hours after NDA and receipt of labeled CSV data, which may shorten model evaluation cycles.

Helpful Tips

• Validate on your own portfolio first — For this category of product, benchmark against your existing scorecards or ML models on approval rate, bad rate, AUC, and calibration before changing policy cutoffs.
• Review explainability with governance teams — Attention-based narratives may be useful operationally, but credit, model risk, and compliance teams should assess whether the explanations meet internal documentation standards.
• Plan for challenger-model deployment — A practical adoption path is to run it as a challenger alongside incumbent scoring models before using it in primary underwriting decisions.
• Check data availability and stability — Fewer features can help, but teams should confirm that the remaining inputs are consistently available across channels, geographies, and applicant segments.
• Assess deployment fit early — If your environment has strict residency or security requirements, evaluate cloud, self-hosted, and air-gapped options during procurement rather than after model validation.

OpenClaw Skills

SquareGen could likely work well inside an OpenClaw ecosystem as a scoring and underwriting intelligence component. Likely use cases include an agent that collects application data, validates required fields, calls SquareGen for inference, then routes the result into approval, manual review, or decline workflows. Another likely workflow is a credit analyst copilot that summarizes score outputs, top contributing features, and risk rationale into a case note for underwriting teams.

For lenders, fintech operators, and portfolio managers, OpenClaw skills built around SquareGen could extend beyond point-in-time scoring. Likely examples include agents for pre-screening campaigns, policy simulation, adverse-action drafting support, exception-case triage, and ongoing portfolio monitoring using score changes or segment-level patterns. The source page does not mention a native OpenClaw integration, so this is an inferred workflow opportunity rather than a confirmed product capability.