AI ASSISTANT / KNOWLEDGE BASE

AI assistants built on your company’s knowledge

The assistant answers using only your materials—in Russian, Kazakh, and other languages, even when the data is complex.

  • Answers using only your data
  • Multiple languages
  • Complex data structures
SAMPLE ANSWERSAI / KNOWLEDGE BASE
  • Question in RussianAnswer from the knowledge base
  • Question in KazakhAnswer from the knowledge base
  • Answer found in the materialsSource
  • No answer in the knowledge baseHand off to a person
  • Question is outside the topicNo answer
Answers are based on your data

Illustration, not a real project

150+projectsdelivered
since 2017yearsin the market
Multipleoutput channels
CHALLENGES

Why company knowledge is not working for your business

The materials exist, but finding answers takes time, while a generic chatbot does not follow your rules.

Policies, instructions, and spreadsheets are scattered across different places. The assistant finds the relevant passage and answers in plain language.

Knowledge search

Time to find an answer: from 15 minutes to 1 minute

Types of assistants we build

The data sources and audiences vary, but the principle is the same: answers come only from your materials.
01

Customer assistant

Answers questions about products, terms, and order statuses.

Works on your website and in messaging apps, handing complex questions to a human agent.

02

Employee assistant

Finds answers in policies, instructions, and the knowledge base.

Helps new employees get up to speed and reduces questions for senior colleagues.

03

Learning assistant

Works with learning materials across subjects, languages, and grades.

As in the Digital Mektep project, it searches across the books and synthesizes the relevant information.

04

Sales assistant

Recommends a product and explains its terms using your descriptions.

Helps the manager and customer reach a decision sooner.

05

Document search

Finds relevant passages across large document collections.

Shows which document the answer came from.

06

Multilingual assistant

Understands questions and answers in any number of languages.

We select a combination of models and translation tools to suit the task.

Did not find your use case? Tell us about the task and we will let you know whether an assistant is a good fit. Discuss your task ↓
SERVICES

What we do: from your materials to a working assistant

We do more than connect a model. First, we organize the materials, then configure the assistant and check its answers.
  1. Data assessment and preparation

    We review the materials, find gaps and duplicates, and bring them into a consistent format.
    Inventory of materials and gaps
  2. Knowledge structure

    We describe how the data is related: sections, languages, versions, and levels.
    Data schema
  3. Configuration and tuning

    We connect an existing model to your data and configure its answers.
    Answers based only on your data
  4. Languages and response tone

    We configure the languages, style, and limits of what the assistant can discuss.
    No limit on the number of languages
  5. Quality checks

    We check answers against a set of review questions before launch.
    Set of evaluation questions
  6. Launch and support

    We connect the assistant to your website or channels, review real questions, and make improvements.
    Support
Read How we work on a project and Data and security in AI projects. Details about data structure and validation are below ↓

TECHNOLOGY

Existing models, your data

We do not train a model from scratch. We choose an existing model and build a solution around your materials.

  • ChatGPT
  • Claude
  • Gemini
  • Qwen
  • DeepSeek
  • TypeScript
  • Next.js
  • Python
  • Node.js

Existing models

We select a model for the task and languages, including ChatGPT, Claude, and Qwen. If cost efficiency matters, we look for affordable options, including Chinese models.

How we choose a model

Search across your data

The assistant finds relevant passages in your materials and answers from them.

Answer with a source reference

Hosting

We host it on our infrastructure or on a server of your choice.

nadvan.kz/hosting

We choose models and a technology stack for each task. Read more about checks in Website security and performance.

PROCESS

4 stages from assessment to launch

Clear steps, timelines, and your role at each stage.
  1. Task and data assessment

    from 5 days
    We review the materials, user questions, and channels, then define what the assistant should and should not do.
    What we need from you: Access to materials and sample questions
  2. Pilot with your data

    from 4 weeks
    We demonstrate the assistant using a portion of your materials.
    What we need from you: A project owner and answer review
  3. Configuration and integrations

    We connect the materials, languages, channels, and systems, then check quality.
    What we need from you: Feedback on answers
  4. Launch and support

    go-live
    We launch, review real questions, and make improvements.
    What we need from you: Approval to launch

We manage tasks in Jira on a dedicated project board, where you have full access. How we work →

We provide timelines and pricing after assessing the data.

OUTCOMES

What you get

01

A working assistant

Available in the selected channels: your website, messaging apps, or internal systems.

02

A prepared knowledge base

Structured materials for the assistant to use.

03

Handoff to a person

Complex questions are passed to a specialist with the conversation history.

04

Question analytics

What people ask, which questions have no answer, and what to add.

05

Documentation and training

How the assistant works and how to update its materials.

PROJECT

Digital Mektep: an assistant for books, subjects, languages, and grades

The task

Gather and provide information from materials across books, subjects, languages, and grades

What we did: We built an assistant that searches across books, subjects, languages, and grades, then synthesizes the relevant information.

PRICING

Pricing depends on your data

There are no fixed plans here: first we assess the data, then prepare an estimate.

Data volume and condition

The amount and format of your materials

The more extensive and scattered the materials, the more work is needed to prepare them.

Languages and structure

The number of languages and data levels

There is no limit on the number of languages. A multi-level structure, as in Digital Mektep, requires more configuration.

Channels and integrations

Where the assistant runs

Website, messaging, CRM, and other system integrations are priced separately.

Pricing is tailored. For smaller projects, the assessment is included in the estimate; for larger projects, we provide an hourly rate after reviewing the project. Model request costs and third-party platform licenses are charged separately. Payment terms are described in How we calculate project cost and payment terms.

Frequently asked questions

Other AI services

If another service is a better fit, we will say so and point you to the right page.