The Ministry of Economy, Ecology and Agriculture of Ukraine digitizes licensing process with Gemma Ukraine uses Gemma to build a more efficient and transparent system for government-issued licenses. ePermit is Ukraine’s online service for applying for permits and licenses, developed by the Ministry of Economy, Ecology ad Agriculture of Ukraine. The service lives within the country’s Diia portal, a digital hub which was created as part of a larger initiative to digitally transform Ukraine and its infrastructure.The Ministry developed ePermit with the goal of simplifying regulations, fostering entrepreneurship, and improving the accessibility and transparency of government services for citizens and businesses. To deliver a more efficient and accurate review process for applications, the Ministry integrated Gemma 3 27B into its workflow.The challengeIt can take a government official hours to manually review a single permit application because they have to read every submitted document and cross reference names, dates, IDs, and more. Not only is the process time-consuming, it can also be prone to human error and subject to inconsistency and biases. The lengthy review process can cause costly mistakes and delays, undermining the Ministry's efficiency.The Ministry also wanted to offer more timely feedback to applicants waiting for approval. Applicants with incorrect or incomplete submissions were left waiting for days, only to be notified of the errors after the manual review was complete, which created frustrating delays and further lengthened the application process.Using AI to improve the efficiency and accuracy of the process was a logical solution, but Ukraine’s legal framework presented some important safety restrictions: All AI systems involved in Ukrainian public services must operate within a closed infrastructure, with no external API calls or cloud dependencies. “ Bringing AI into government services is complex. We must work within strict legal and security requirements, keeping all operations in a fully closed environment with no external APIs or cloud. Gemma gave us the rare balance of strong reasoning and secure on-premise deployment. Oleksandr Tsybort, CDTO and Deputy Minister of Economy, Environment and Agriculture of Ukraine The solutionTo address that set of challenges, the Ministry built a system that analyzes document information extracted from scanned images that are submitted, combines it with data retrieved from government registries, and recommends a decision to the government official assigned the application.To decide which LLM to base their model on, the developers at the Ministry tested several open ones to find the best suited to their use case. The team discovered that Gemma 3 27B consistently outperformed the other models they tested when it came to accuracy, logical consistency, and auditability, especially in structured decision-making workflows. “Gemma’s performance and stability made it the most suitable choice,” said AI Advisor Dmytro Voitekh. “An added benefit of Gemma is its ability to be efficiently deployed on-premise, thanks to its relatively compact size. Gemma's model line is one of the best open models suitable for most cases, including Ukrainian language and context — especially important while Ukraine continues developing its own national Ukrainian LLM.”The Ministry’s solution uses Qwen 2.5 7B VL to extract the text data from scanned documents, and Gemma 3 27B to ensure the application is complete, the information is consistent, and that it meets certain requirements, such as the minimum age and education of applicants. The model then reasons and decides on one of three recommendations to give to the official assigned the application: issue the license, reject the application, or conduct further manual review. The model is now deployed on-premise using VLLM, running on NVIDIA H100 GPUs in a Ukrainian datacenter. “ Gemma features a rare balance: strong reasoning capabilities, transparent open-source licensing, and lightweight enough for secure on-prem deployments. Dmytro Voitekh, AI Advisor at DM Tsybort The impactThis new model is used within ePermit for reviewing veterinarian licenses, but the team has already determined that the AI is cutting processing time dramatically and is running without latency issues. It’s reducing rejection rates, resubmissions, and workload for government officials.Gemma also offers the developers at the Ministry additional information they can use to further improve the model’s performance and capitalize on their successes. Gemma’s decisions are auditable, meaning all of Gemma’s inputs, outputs, and rules are logged, enabling the team to make changes and improvements until they achieve their desired result. “ We're excited to see how open models like Gemma can empower public institutions to adopt AI responsibly, within legal, ethical, and infrastructural boundaries. We believe collaboration between governments and tech leaders like Google are key to building trustworthy and sovereign digital systems. Oleksandr Tsybort, CDTO and Deputy Minister of Economy, Environment and Agriculture of Ukraine What’s nextAs the development team works to expand use of the model to other license types beyond just veternarian licenses, they are also fine-tuning it to improve its understanding of the Ukrainian language and its document-specific terms. This will involve building an evaluation framework to further refine the performance and accuracy of the model. The team is also experimenting with a unified multimodal approach, fine-tuning Gemma 3 on their document scans dataset to enhance visual document understanding.Voitekh recommends Gemma for developers who are building closed systems: “If you’re working in a domain with complex logic or constraints—like government, legal, or regulated industries—it’s absolutely worth testing.”