Sun Finance automates ID extraction and fraud detection with generative AI on AWS

This article details how Sun Finance, a Latvian fintech, used AWS generative AI services to automate identity document extraction and fraud detection, significantly improving accuracy and reducing costs.

• Improved ID extraction accuracy from 79.7% to 90.8% using multi-tier OCR approach
• Reduced per-document costs by 91% compared to previous solution
• Cut processing time from up to 20 hours to under 5 seconds per document
• Combined Amazon Textract, Amazon Rekognition, and Claude for intelligent document structuring
• Built serverless fraud detection using visual pattern analysis and background similarity search
• Achieved 81% fraud detection accuracy with 59% recall and 83% specificity
• Projected manual review workload reduction from 60% to 30% of applications
• Solution deployed in 35 business days from technical handover to production
• Architecture uses AWS Lambda, Step Functions, Bedrock, and S3 Vectors for scalability

Sun Finance demonstrates how combining specialized OCR tools with LLMs outperforms single-tool approaches for document processing, enabling rapid expansion into lower-value loan markets.