Biorg.Documents

2026: Handwritten Font Recognition

The neural network of the Biorg company sees handwritten symbols, like a person: it perceives all information in the image at once, and not in stages. This approach provides accuracy at a level of 80% even for recognizing "noisy" handwritten data in the passport registration stamp, without video cards (GPU). The technology can be scaled for digitization and other types of documents. Biorg announced this on March 13, 2026.

At the first stage, the neural network was taught to understand handwritten text in simple lines, and then "read" complex and "noisy" text, where handwritten characters alternate with printed ones. As a result, the model detects character features in 3-4 seconds, understands their vector and decodes it. Thus, on any handwritten fragment, AI recognizes data with completeness and accuracy at the level of 80 percent, regardless of the type of document.

The neural network was trained on open arrays and its own Biorg dataset using real and synthetic images - in total, more than 30 thousand fragments of handwritten data were required. Additional training will make it possible to raise the accuracy indicator above 90%.

The technology is relevant not only for recognizing registration stamps in the passport, but also for automatically digitizing handwritten fragments of certificates, work books, military tickets and other forms in companies and business processes banks insurance other corporate customers. The next step is to transfer the revision from SaaS the -version to the "boxed" version of our " ON Biorg.Documents." High-quality handwriting recognition is another milestone in the path of continuous improvements that we adhere to, "said Biorg CEO. Ruslan Aligadzhiev

The problem of recognizing handwritten data traditionally lies in the absence of a single style of writing letters. In the passport registration stamp, the situation is aggravated by the presence of "noise": printed text is mixed with handwriting, the distance between lines varies, there is no geometric sequence, there is no uniformity of writing addresses - all this dramatically affects the quality of recognition.

Unlike classical approaches, where you need to determine the type of document, turn it, turn words into lines and recognize them, the Biorg neural network perceives the image as an end-to-end data set.

2024: Government's refusal to allocate funds for product creation

In mid-January 2024, it became known that two projects for the development of software with optical document recognition (OCR) functions were excluded from the New System-Wide Software roadmap. We are talking about the creation of analogues of the FineReader series of ABBYY, which left the Russian market due to the current geopolitical situation. The total cost of rejected projects was supposed to exceed 0.5 billion rubles.

According to CNews, the state refused to finance Biorg.Docs and Easydoc products. The first, as expected, Biorg will create by order of Rostelecom. It was planned to develop a cross-platform solution for digitizing documents and recognizing images in the interests of government and corporate customers. In turn, ITFB's Easydoc (iT & B Distribution) was positioned as an intelligent document processing platform that extracts data based on artificial intelligence technologies. The costs of creating Easydoc alone were estimated at 512 million rubles.

Two projects for the development of software with optical document recognition functions are excluded from the roadmap "New system-wide software"

Ilya Verigin, director of interaction with state authorities at Biorg, said that the company, together with Rostelecom, approached the approval of the Biorg.Docs project within the framework of the development competence center (CDC), but then uncertainties arose. The fact is that Prime Minister Mikhail Mishustin instructed to prioritize industry information systems that are critical for enterprise management. As a result, the CDC reoriented to industrial software, and Rostelecom withdrew its order for Biorg.Documents.

In such a situation, we decided to get away from the idea of ​ ​ ABBYY replacement and focus on the original concept of our platform, "says Verigin.

At the same time, the ITFB reported that against the background of the state's refusal to finance Easydoc, the company intends to develop this product at its own expense and at the expense of the Innovation Assistance Fund.^[1]

Notes