Eurosoft

Private LLM and RAG on your own servers in Uzbekistan

Eurosoft builds private AI that runs fully on your servers: open LLMs, search over company documents with source links, access per department and no external API.

Updated

In shortEurosoft is a software and AI company in Tashkent, registered in 2019 and a resident of IT Park, with 120+ projects for 55+ clients in 7+ countries. We build private AI platforms that run fully inside your own data center: open large language models on your GPUs, search over company documents with a link to the source of every answer, and access rules per department. Our Enterprise RAG Platform runs this way at Samsung and Incheon International Airport, with no external API and no internet connection.

Company
Eurosoft LLC, Tashkent, Uzbekistan
Working since
2019, resident of IT Park Uzbekistan
Track record
120+ projects for 55+ clients in 7+ countries
Team
30+ full-time employees and 60+ specialists in the network
Clients include
Samsung, Incheon International Airport, ministries of Uzbekistan, World Bank and EU projects
Delivery
Cloud or fully on-premise; you receive the source code
Languages
Uzbek, Russian, English
Contract
Staged delivery, payment after each accepted stage, written warranty

When private AI is the right choice

  • Your documents include contracts, customer or patient data, financial records or anything that must not leave the organization.
  • You are a public body or a regulated company and need to show where data is stored and who saw it.
  • Since 21 January 2026, Law O'RQ-1115 of Uzbekistan requires that a decision affecting a person's rights is not based only on an AI system's conclusion, so answers must be checkable: sources, review and logs.
  • For public information such as prices and opening hours, a cloud assistant is usually enough and costs less.

What the platform does

  • Reads PDF, Office and HWP files and scans: OCR and a vision model extract text, tables and images.
  • Hybrid search: keyword and vector search with reranking, so the answer is built from the right passages.
  • Answers with links to the exact source document, so a person can check them in seconds.
  • A separate knowledge base per department; access is checked before every answer.
  • Connectors for Google Drive, Amazon S3, network storage (NAS) and Jira, and agents for reports and multi-step tasks.
  • A processing queue and health monitoring for large uploads of 1,000+ documents.

Models and hardware

  • Open models run on the client's GPUs: GPT-OSS 120B, Llama 70B, Qwen 32B and Gemma, with Infinity Parser for documents and bge-m3 for search.
  • The model is chosen for the task and the hardware you have: a smaller model on one server for a pilot, a large model on a GPU cluster for the whole company.
  • Nothing is sent to an external API; the platform works with no internet connection at all.

Where it runs

  • Enterprise RAG Platform runs 100% offline, on-premise, at Samsung and Incheon International Airport, including a 120B-parameter model on the client's GPU cluster.
  • In Uzbekistan the platform is offered as SarhadAI.

How a project runs

  • A short discovery call with the people who will use the system, then a written specification split into stages with dates.
  • Each stage is delivered, shown to you and accepted before the next one starts; payment follows acceptance.
  • You get the source code, documentation and access to the repository. After launch there is a written warranty period, then support on agreed terms.
  • The system runs in the cloud or fully on your own servers, whichever your data and rules require.

Case studies

Questions and answers

What is a private LLM?
A large language model that runs on your own servers instead of a provider's cloud. Your questions and documents stay inside your network, and you decide who can use it and what it can read.
What is RAG?
Retrieval-augmented generation: before answering, the system searches your documents and gives the model only the relevant passages, so the answer is based on your data and shows its source.
Which companies in Uzbekistan build private LLM and RAG platforms?
Eurosoft is one of them: its Enterprise RAG Platform runs fully offline at Samsung and Incheon International Airport, including a 120B-parameter model on the client's GPU cluster, and is offered in Uzbekistan as SarhadAI.
Do we need our own GPU servers?
For a fully private platform, yes. A pilot can start on a single GPU server with a smaller open model; we size the hardware after a short review of your documents, users and response-time needs.
Can it answer in Uzbek and Russian?
Current open models handle Uzbek and Russian; we test them on your real questions and documents before launch and pick the model that answers best in your languages.

Guides

Have a project in mind?

Tell us what you are building and where it has to run. We will come back with a proposed next step.