NVIDIA named Tajikistan an example of AI implementation in government services

In Tajikistan, the Ijozat AI platform was launched, which uses artificial intelligence to process applications for industrial licenses. Previously, reviewing one application took from 14 to 30 days; now the process takes about 90 seconds. NVIDIA published the project story in the Success Stories section of its official website.

The platform was developed for the Ministry of Industry and New Technologies of Tajikistan. It automatically recognizes documents, determines their type, checks data against state registries, and identifies potential violations. The SoroLLM language model analyzes regulatory requirements, answers applicants' questions, and generates brief summaries for decision-making staff.

The system processes documents in Tajik, Russian, and English. Each decision is accompanied by a verification history. Thanks to automation, the volume of repetitive work for employees has decreased by 50-60%.

The platform is based on SoroLLM — Tajikistan's first national large language model, presented in June 2025 with the participation of President Emomali Rahmon. The zehnlab.ai team fine-tuned the model on a corpus of Tajik language texts. According to NVIDIA, the Tajik language accounts for less than 0.01% of texts in the Common Crawl archive, one of the largest open collections of internet data.

For additional safety verification of responses, developers created Soro Guard 1B — a model that detects unsafe content in Tajik, Russian, and English.

All computations and personal data processing take place within the country. The platform operates on a Supermicro server with eight NVIDIA H200 graphics processing units in the sovereign data center darya.ai in Darvaz. The infrastructure uses NVIDIA CUDA technology. The language models were developed by the zehnlab.ai team, and the Ijozat AI application was created by Epsilon3.ai.

According to NVIDIA, more than 130 industrial licenses have already been issued through the platform. Over the next 12 months, it is planned to process another 1,500 applications. In the future, similar automation processes are planned to be implemented in another 70 state institutions in Tajikistan.

In the long term, the technologies are planned to be used for fraud detection, modeling the consequences of government decisions, and issuing licenses in the healthcare sector.

Source: NVIDIA