Perne
A custom ML model for your task: illustration 1
Data and AI analytics

A custom ML model for your task

We work as a team and start by clarifying the task: what needs to be predicted or detected, what data you have, and how you will judge that the model is useful. Then we clean the data, choose suitable algorithms from scikit-learn and XGBoost, train several options and compare them on quality metrics. The best one is packed into a script or a simple API that can be connected to your site, CRM or spreadsheet. A model depends on its data: if the data is scarce or messy, the result will be more modest, and we will say so honestly before the work starts.

from 150,000 ₸ (≈ $300) one-off

timeline: 14–30 working days

  • Analysis of the task and agreement on the metric we will use to judge the model
  • Checking and cleaning your data: gaps, duplicates, outliers
  • Training and comparing 2-4 models from scikit-learn and XGBoost

Order

Message on WhatsApp
TelegramCall

The service name is added to the message for you.

This is a starting price; we give the final amount before work starts. USD is approximate.

What is included

  • Analysis of the task and agreement on the metric we will use to judge the model
  • Checking and cleaning your data: gaps, duplicates, outliers
  • Training and comparing 2-4 models from scikit-learn and XGBoost
  • A report with quality metrics and a plain explanation of what the model relies on
  • A script or simple API to run the model on new data
  • Instructions for running and updating the model on your data

Who it is for

For companies that already collect data on sales, leads, customers or equipment and want a forecast or automatic classification. It is not a fit if you have almost no data or if you need accuracy guaranteed by contract: machine learning gives probabilities, not guaranteed answers.

Not included

To avoid surprises, here is what the base price does not cover and what is discussed separately:

  • Collecting data from scratch and labelling large volumes by hand: a separate job with its own estimate
  • A guarantee of a specific accuracy percentage or of money results
  • Cloud computing and server rental for training: billed separately by actual use

FAQ

After payment the code, the script and the trained model are handed over to you, and you can use them in your business. The scikit-learn and XGBoost libraries are distributed under their own open licences, and we follow them. If the data contains other people's materials or personal data, the party that provided it is responsible for its legality, so we check this at the start.

The price includes two rounds of revisions after the first model version: changing features, trying another algorithm or adding new metrics to the report. If the task itself changes or new data is needed, that is a new scope of work, which we agree in advance.

Usually 14 to 30 days from the moment we receive the data and agree on the task. Most of the time goes into preparing the data and comparing options, so the timing depends on how clean your files are. We will name an exact date after the first look at the data.

A description of the task and how you will use the result, an export of the data in a spreadsheet or database with clear column names, and a note on what the fields mean and how the data was collected. It helps to say what counts as a correct answer, for example whether the customer eventually bought. If we have no access to your system, files are enough.

No, we do not promise a specific accuracy figure in advance. After training we show metrics on data the model has not seen, so you can judge whether the result is good enough. If quality is low, we will say honestly why, for example because there is too little data or the features are weakly linked to the outcome.

We hand over a script that reads new data from a file or database and returns a forecast, and a simple Python API if it needs to be connected to your site or CRM. The instructions describe how to run it and retrain it on fresh data. You choose the server and where the model is stored, and we can help with setup by agreement.

The model is trained on past data, and its accuracy may drop if customer behaviour or market conditions change. The instructions describe the signs of such a shift and how to retrain. Regular support and repeated training are paid separately by agreement.

You may also need

Demand forecast

Demand forecast

A sales forecast by product for weeks or months ahead, with an error estimate and a clear table for purchasing.

from 60,000 ₸ (≈ $120) one-off

Timeline: timeline: 7–14 working days

Computer vision: visitor counting, queues, product recognition

Computer vision: visitor counting, queues, product recognition

We set up video analysis from your cameras: visitor counting, queue monitoring and product recognition on shelves, with no identification of people.

from 200,000 ₸ (≈ $400) one-off

Timeline: timeline: 14–30 working days

Shall we discuss your project?

The product name and a link to this page are filled into the WhatsApp message for you.

Message on WhatsApp
TelegramCall

Or send a request

We will reply in the channel you choose: WhatsApp, Telegram, phone or email. Your contact is used only to reply.

Service
How to reply

from 150,000 ₸ (≈ $300) one-off

Message on WhatsAppTelegramCall