ABBYY gives old-school OCR a job in the AI pipeline

ABBYY has introduced a new tool called FineParser that packages its FineReader OCR engine for use in AI systems. This tool is designed to turn documents into structured text that can be easily processed by modern generative AI language models. FineParser runs in a Docker container on a CPU, without the need for a GPU, and aims to preserve the layout of a document while extracting its contents.

The company describes FineParser's approach as "deterministic AI," meaning it extracts text and document structure rather than generating plausible versions of them. This makes the output less predictable and more suitable for use with generative AI systems. FineParser handles documents in multiple languages, recognizing columns, headings, and tables even without borders, and can process handwriting as well.

ABBYY offers FineParser as a self-hosted tool, meaning it can be run on a user's own server or device. This approach allows for greater control and flexibility, especially for those who may not have access to a GPU. The tool also provides a free tier, allowing for 1,000 pages per month for one year, and subscription tiers for more extensive use. For companies that prefer a completely offline solution, there is an Enterprise plan.

Developers can integrate FineParser into their own products using a programmable machine-learning framework called NeoML, which is open source and available on GitHub. The tool can be accessed through a REST API, with output available in plain text, JSON, or DocLang format, a compact format designed for easier input into language models.

ABBYY believes that the layout-preserving approach of its OCR technology, running on a CPU, still holds value in the AI pipeline, demonstrating that sometimes the old-fashioned methods can still be useful.