Skip to main content
sətir

Data and AI

Chat assistants in Azerbaijani and Russian, reading scanned documents, sales and load forecasts. On your data and your servers.

Price
from ₼14,000
Length
3–14 weeks
Team
2–4 people
Contact centre agents with headsets at their desks

What we do

Every project starts with a short proof of concept: in 3 to 4 weeks, a working sample on your real data and a test set that measures its accuracy. If the result is not good enough, we stop and say so plainly. Where we can, we run the models on your servers, and client data does not go to an outside service.

  • Chat assistants

    Answers from your tariffs, rules and documents, with the source under every answer and hand-over to an agent.

  • Document reading

    Fields pulled from scans: applications, contracts, certificates, in Cyrillic and Latin script.

  • Forecasts

    Sales, warehouse stock, contact centre load; by week and by day.

  • Analytics

    Reports and dashboards on ClickHouse, without loading the operational database.

  • Checks and control

    A test set, a weekly quality report, a log of wrong answers.

How the work goes

  1. 3–⁠4 weeks

    Proof of concept

    A sample on real data, a test set, an accuracy figure.

  2. 6–⁠8 weeks

    Pilot

    In one department or channel, with limited users.

  3. 3–⁠6 weeks

    Roll-out

    All channels, integrations, a load test.

  4. ongoing

    Control

    Quality reports, updates with new documents.

Price

Proof of concept, 3 to 4 weeks
₼14,000–⁠19,000
Pilot, 8 to 10 weeks
₼45,000–⁠60,000
Data and ML engineer, hourly
₼80
GPU server
yours or the data centre’s, billed separately

Prices without VAT (18%), at the rates in force since 1 January 2026.

What you have at the end

  • A test set that measures accuracy, with the results
  • The model and code running on your servers
  • A panel for marking wrong answers
  • A diagram of where the data goes

Technologies

  • Python
  • PyTorch
  • ClickHouse
  • PostgreSQL

Projects with this service

Questions

What if the assistant gives a wrong answer?

It answers only from your documents and shows the source. When unsure, it calls an agent. We find mistakes through the weekly panel and add them to the test set.

How well does it understand Azerbaijani?

We measure that in the proof of concept with your clients’ real questions. On our last banking project the test set had 1,900 questions.

Estimate your project

Write one sentence or pick the screens, integrations and deadline: the team, the weeks and a price range appear at once.

Go to the estimatorBook a call
Call usMessage on WhatsAppSupport ticketEstimate