Private AI and the systems that run a business.
We build the AI layer and the software around it, and deploy it where your data has to stay: in the cloud or fully on your own servers.
Services
One team for the AI layer and the systems around it.
An assistant is only useful when it is connected to the CRM, the warehouse, the payments and the people who use them. We build both sides.
- 01Core focus
AI and LLM platforms
RAG, private LLMs and AI agents, deployed in the cloud or fully on-premise.
- 02Business systems
CRM
Sales, customers and service in one system, built around how your team works.
- 03Business systems
ERP
Finance, stock and operations, including warehouse flows with scanner apps.
- 04iOS and Android
Mobile apps
Customer ordering apps and internal tools for staff in the field.
- 05Sites and portals
Web
Websites, portals and dashboards for companies and government bodies.
- 06Online sales
E-commerce
Stores, ordering and payments, from product catalogue to delivery.
- 07Messaging
Telegram bots
Sales and support bots that meet customers where they already are.
- 08Advisory
IT consulting
Architecture reviews and audits before you commit to a build.
The challenge
Most AI projects stall between the demo and the data center.
A chatbot on a public API is easy to show and hard to deploy. Real organisations have confidential documents, legacy file formats, and decisions that someone has to answer for. We build for those constraints from the first day.
Data that cannot leave the building
Many enterprises and public bodies cannot send documents to an external AI API. Our RAG platform runs fully offline on the client's own GPUs, with no internet connection at all.
Documents that machines cannot read
Company knowledge sits in scanned PDFs, Office files and HWP. We extract text, tables and images with OCR and vision models before anything is indexed.
AI with nobody accountable
A model's output is not a decision. In AI Psixolog a psychologist confirms every flagged case, referrals need consent, and each AI function can be switched off.
Architecture
Every model runs inside the client's data center.
This is how the Enterprise RAG Platform is deployed at Samsung and Incheon International Airport: documents in, answers with sources out, and nothing crosses the network boundary.
Client data center
Works fully offline, no internet needed
Ingestion
Documents
PDF, Office, HWP
Parsing
OCR + vision model
Index
Keyword + vector
Knowledge bases
Separate per team
Each question
Question
Korean or English
Access check
Own team only
Retrieve
Search + rerank
Answer
With source links
GPU cluster all models run here
- GPT-OSS 120B
- Llama 70B
- Qwen 32B
- Gemma
- Infinity Parser
- bge-m3
Document parsing
OCR, tables and images from PDF, Office, HWP and scans.
Hybrid search
Keyword and vector search with reranking.
Private LLMs
Open models on the client's GPUs.
AI agents
Reports and multi-step tasks.
Access control
A separate knowledge base per department.
Reliability
Processing queue and health monitoring.
Technology
The stack behind the products.
Open models where the data has to stay private, managed services where speed matters more, and standard engineering tools around both.
- AI models
- Llama
- Qwen
- Gemma
- bge-m3
- Ollama
- GPT-OSS 120B
- Model training
- VLM: vision-language models
- LLM: large language models
- TinyLM: small language models
- MLOps: training and deployment
- Cloud and agents
- AWS
- Google Cloud
- n8n
- MAF
- Vercel
- MCP
- Engineering
- Python
- FastAPI
- Next.js
- React
- Vite
- Data and search
- Elasticsearch
- PostgreSQL
- Neo4j
- Amazon S3
- Integrations
- ERP systems
- Telegram
- Google Maps
- Google Drive
- Jira
- OpenAI GPT-4o-mini and Whisper
- KG Inicis payments
Frequently asked
What teams ask us before they start.
What is Eurosoft?
Eurosoft (Eurosoft LLC) is a software company in Tashkent, Uzbekistan, registered on 12 January 2019 and a resident of IT Park Uzbekistan. It builds private LLM and RAG platforms, AI assistants, CRM and ERP systems, mobile apps, websites and e-commerce. Its Enterprise RAG Platform runs on-premise at Samsung and Incheon International Airport, and its AI Psixolog platform is used by 350,000+ users in 250+ schools and technical colleges in Uzbekistan.
Where is Eurosoft based, and where does it work?
Eurosoft's office is at Zafar Diyor Street 119A, Mirzo Ulugbek district, Tashkent 100007, Uzbekistan. It delivers projects in Uzbekistan (including Namangan), South Korea (Seoul and Incheon) and the United Arab Emirates (Dubai).
Who are Eurosoft's clients and partners?
Clients include Samsung, Incheon International Airport, Intel, Lenovo, Flagshop, Mr. Papa and SkyBlue Group, and government bodies in Uzbekistan such as the Ministries of Agriculture, Water Resources, Health, Justice and Digital Technologies. In October 2026 Eurosoft signed a partnership agreement with Azati, a software engineering and AI company with 300+ engineers delivering in the EU and the US.
Can the AI platform run entirely on our own servers?
Yes. The Enterprise RAG Platform is built for it. At Samsung and Incheon International Airport it runs 100% offline inside the client's data center, with no external API and no internet connection. All models, including a 120B-parameter one, run on the client's GPU cluster.
Which models do you deploy?
On-premise we run open models: GPT-OSS 120B, Llama 70B, Qwen 32B and Gemma, with bge-m3 for search and Infinity Parser for document parsing. For cloud products such as Viji AI we use OpenAI GPT-4o-mini and Whisper. We also train vision-language, large and small language models.
What documents and data sources can it read?
PDF, Office and HWP files, including scans. Parsing uses OCR and a vision model to keep tables and images. Connectors sync from Google Drive, S3-compatible storage, network storage (NAS) and Jira, and the platform handles 1,000+ documents per upload.
How do you keep each department's data separate?
Each department gets its own knowledge base. Before any retrieval the platform checks access, so people only get answers from their own team's documents, and every answer comes with links to its sources.
Do you only build AI products?
No. We also build CRM and ERP systems, mobile apps for iOS and Android, websites and portals, e-commerce, Telegram bots, and we do IT consulting. Examples include Flagshop in Seoul, O'zbegim supplying, the SUIS school platform, Kissa Finance and the AMCHAM Uzbekistan website.
Which languages do your products support?
It depends on the project. The RAG platform answers in Korean or English. AI Psixolog works in three languages. O'zbegim supplying ships in Uzbek (Latin), Uzbek (Cyrillic) and Russian.
How do we start working with Eurosoft?
Send us a message through the form below or write to info@eurosoft.uz. Tell us what you are building, what data and systems are involved, and where it has to run. We will come back to you with a proposed next step.
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.