Ukrainian AI platform Lapathoniia has integrated the Rukopys OCR model for recognizing Ukrainian handwritten documents. The tool is designed for rapid digitization and search in historical and administrative papers — from state archives to school notebooks — and its cloud infrastructure is deployed in Ukraine.
Ukrainian AI platform Lapathoniia has integrated the Rukopys OCR model for recognizing Ukrainian handwritten documents. The tool is designed for rapid digitization and search in historical and administrative papers — from state archives to school notebooks — and its cloud infrastructure is deployed in Ukraine.
This was reported on the blog of the company De Novo, which provides cloud capacity for the project.
The service is based on the compact Rukopys-OCR-4B model, created by retraining Qwen3.5-4B and published under the Apache 2.0 open license. According to the developers, in internal testing on a closed dataset, the domestic model achieved an overall quality of 0.9137 versus 0.7838 in Gemini-3 Flash, making 59–64% fewer errors when reading text on pages and in individual blocks.
The model processes the document in a single pass: it automatically detects text areas, their types and coordinates. The web interface supports uploading JPEG, PNG, WebP, TIFF and multi-page PDF files. Users can edit the recognized text, change the boundaries and block types, and export the result to TXT, JSON or searchable PDF formats.
In addition to the web interface, the service is available via an API that is compatible with the OpenAI format. Testing is free for private users.
The estimated cost of processing large volumes of documents is $0.03 per page in regular mode and $0.018 per page in batch mode. Several organizations have already received access to the system, including the Scientific and Technical Library of Lviv Polytechnic, the First Medical Association of Lviv, the Lviv National Scientific Library named after V. Stefanyk, and the Lviv Museum of the History of Religion.
The basis for creating the service was the Handwritten to Data handwriting recognition competition, which AI HOUSE held jointly with the Ministry of Digital Affairs, the Ministry of Economy, and UCU in April-July 2026.
The Lapathoniia team chose as a basis the model of the Japanese developer Tetsuya Fujimura, who took third place and donated his prize fund to support Ukraine. The updated RUKOPYS dataset is publicly available on Hugging Face under the CC BY 4.0 license.
