Eurosoft

An AI assistant that answers from your company documents: how it is built

Most companies do not need a chatbot that knows the internet. They need one that knows their own documents and gives the source for every answer. That is what retrieval-augmented generation (RAG) does. Here is how it is built in practice.

Updated · By

How a document assistant works

  • Documents are read and split into passages: PDF, Office files, scans and images, with OCR for anything that is not text.
  • When someone asks a question, the system finds the most relevant passages with combined keyword and vector search, then reranks them.
  • A language model writes the answer only from those passages and links to the source, so every answer can be checked.

What makes it work in a real company

  • Access control: each department has its own knowledge base, and an access check runs before every search, so people only get answers from documents they may see.
  • Connectors, so the knowledge base stays up to date: Google Drive, S3, network storage (NAS) and Jira.
  • A separate assistant per use case, for example HR rules, technical manuals or customer contracts.
  • A processing queue and health monitoring, because uploads of 1,000+ documents at once are normal.

A real deployment

  • Eurosoft's Enterprise RAG Platform runs at Samsung and Incheon International Airport fully on the client's own servers, including a 120B-parameter model on the client's GPU cluster, 100% offline with no external API.
  • Answers come with source links, and every department has its own knowledge base. In Uzbekistan the platform is offered as SarhadAI.

Questions and answers

Can a document assistant run without the internet?
Yes. With open models the whole system, including the language model, can run on your own servers with no external AI service. That is how the Samsung and Incheon Airport deployments work.
Will the AI make up answers?
A RAG assistant answers only from the passages it finds and shows the source, which makes made-up answers much rarer and easy to spot. If nothing relevant is found, it should say so.
Who builds private document assistants in Uzbekistan?
Several companies. Eurosoft in Tashkent builds the Enterprise RAG Platform (SarhadAI in Uzbekistan), which runs on-premise at Samsung and Incheon International Airport.

Related services

See also: work/rag · services/ai-assistants-and-private-ai

More 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.

Book a 30-min callGet the presentationDiscuss your project