Agile Soda Launches Agentic OCR Platform with 98% Document Classification Accuracy

Gate News message, April 17 — Agile Soda unveiled Agentic OCR, an AI-powered document automation platform that requires no pre-training and can be deployed immediately. The platform, built on large language models (LLM) and vision-language models (VLM), aims to transform the traditional OCR market by eliminating the need for document-type-specific model retraining.

Unlike conventional OCR systems that require weeks to months of retraining for new document types, Agentic OCR allows users to define extraction fields—such as name, contract date, and amount—and the AI automatically recognizes, classifies, and extracts the information. Users can upload sample documents to auto-generate classification and extraction templates, with same-day deployment requiring no code or training data. In validation tests across 20 document types, the platform achieved 98% document classification accuracy and over 95% information extraction accuracy.

The platform incorporates active learning, improving performance as it is used. When users correct extraction errors, the system automatically incorporates the corrections, continuously enhancing accuracy without additional training—yielding 5–15% performance improvements. Agile Soda has completed performance validation with major domestic banks, insurance companies, and card issuers and is currently rolling out deployments. Future plans include expanding to RAG-based chatbots, document comparison analysis, and workflow automation as a comprehensive document agent platform.