The European Patent Office accelerates innovation with Mistral
European Patent OfficeMistral AI OCRThe European Patent Office partnered with Mistral AI on a three-month proof of concept to automate patent document transcription. A 1B-parameter OCR model was fine-tuned on over 150,000 PDFs (50,000 unique patents) and built into a pipeline converting complex patent documents (with formulas, chemical structures, tables, and figures across multiple European languages) from Markdown to HTML to ST36-compliant XML, deployed on-premises. The system achieves 400,000 pages/day throughput with a character recognition error rate below 1%, cutting lead time from up to 5 days down to a few minutes.
Overview
The European Patent Office partnered with Mistral AI on a three-month proof of concept to automate patent document transcription. A 1B-parameter OCR model was fine-tuned on over 150,000 PDFs (50,000 unique patents) and built into a pipeline converting complex patent documents (with formulas, chemical structures, tables, and figures across multiple European languages) from Markdown to HTML to ST36-compliant XML, deployed on-premises. The system achieves 400,000 pages/day throughput with a character recognition error rate below 1%, cutting lead time from up to 5 days down to a few minutes.
This entry has 14 published fields tied to exact passages in an immutable source capture.
Inspect the highlighted sourceThe challenge
Patent documents frequently embed mathematical formulas, chemical structures, detailed images, and intricate tables within text across multiple European languages, elements that are notoriously difficult for standard digital tools to interpret, often leading to incomplete or inaccurate data structuring.
The solution
The EPO partnered with Mistral AI on a three-month proof of concept, deploying a 1B OCR model fine-tuned on over 150,000 PDFs representing 50,000 unique patents, built into a pipeline that converts complex patent documents from Markdown to HTML and finally into the required ST36 XML format, ready for on-premises deployment.
Reported business value
The system achieved a throughput of 400,000 pages transcribed per day while maintaining a character recognition error rate of less than 1%, cutting lead time from up to 5 days down to a few minutes.
Sources
Open any source and check the claim yourself — that is the point of the register.
This record was researched and written with AI assistance, and its claims were checked against the sources above. (EU AI Act art. 50 transparency notice.)
Other government & public sector entries in the register.
GovTech unlocks data insights to improve nationwide services with Databricks
Singapore's Government Technology Agency (GovTech) migrated from an on-premises dashboarding system to the Databricks Data + AI Platform on AWS with Unity Catalog and Delta Lake, cutting dashboard creation time from 90 to 30 days, democratizing data across 50% of corporate divisions in the first year, and saving 8,000 labor hours annually.
GovTech· SingaporeDelta Lake · Lakeflow Connect · Lakeflow Jobs +1Estonia rolls out Bürokratt, an AI-guided virtual assistant network for public services
Bürokratt is a network of chatbots deployed on Estonian public sector institutions' websites, letting people obtain information from institutions and use public and information services via virtual assistants. It is a state-created, AI-based digital assistant that helps institutions deliver modern, efficient, around-the-clock customer service using large language models.
City of Helsinki pilots watsonx.ai for multilingual customer care chatbot translation
The City of Helsinki collaborated with IBM Client Engineering and IBM Consulting on a Proof of Value testing whether IBM watsonx.ai could translate foreign-language questions into Finnish for its customer care chatbot, which supports Finnish, English, and Swedish. Watsonx.ai scored 0.90 on the COMET machine-translation metric for English-to-Finnish translation (versus 0.91 for the incumbent system) and 0.87 for Swedish-to-Finnish (versus 0.86), performing on par with the city's existing translation solution.
City of Helsinki· FinlandIBM watsonx.ai
Madrid builds VisitMadridGPT on Azure OpenAI to answer tourists in 95+ languages
The City of Madrid, with partner iUrban, built the VisitMadridGPT virtual assistant on Microsoft Azure OpenAI Service to give tourists accurate, real-time, personalized information in more than 95 languages, drawing from the city's official tourism content. Users average 2.8 questions and spend about 4 minutes 18 seconds per session with a low bounce rate, and the city analyzes assistant conversations to identify popular topics and refine its digital marketing strategy.
Madrid Destino / Madrid City Council· SpainAzure OpenAI ServiceWas this helpful?
Your feedback helps us improve our use case database