nbm-it@mail.ru
Logo NBM-IT - website development studio
+7 (911) 814-86-78

Development of AI models in NBM-IT

We create and train models using company data: for text and image analysis, forecasting, scoring and recommendations.

We test the hypothesis on a PoC, compare the quality with ready-made solutions and bring the selected option to production.

Discuss the AI model

When you need your own AI model

A ready-made API is not enough when it's important your data, metrics and decision rules.
First, we check whether the problem really needs to be solved by a separate model. If a ready-made service is cheaper and more accurate, we honestly offer integration instead of developing from scratch.
Learning from your data
We prepare the dataset, clean, mark and control the quality of the data.
Verifiable Accuracy
We fix the metrics in advance and compare the model with the basic solution.
PoC, MVP and production
We do not transfer the experiment to a production environment without load and business tests.
Integration into processes
We connect the model to CRM, ERP, 1C, website, application or internal API.
Data control
We separate access and choose a cloud, local or hybrid circuit.
MLOps after launch
We monitor the quality, data drift, versions and retraining of the model.
PoCHypothesis testing
MVPTest on a real process
APIIntegration with systems
MLOpsPost-launch control

How it goes AI model development

Each stage ends with a verifiable result: requirements, dataset, metrics, working model or integration.

1

Formulating the task and metrics

We determine the business outcome, limitations, cost of error and criteria by which the model can be run.
2

Checking and preparing data

We analyze the volume and quality of data, eliminate leaks, and create training and control samples.
3

Creating a PoC

We compare several approaches and ready-made models so as not to waste budget on architecture without a proven effect.
4

We bring the model to production

We optimize the speed and cost of inference, design the API, access roles, logs and error handling.
5

We integrate into business systems

We connect the model to the necessary sources and interfaces, conduct acceptance and load tests.
6

We monitor and retrain

We control the quality using new data, model versions and safe rollback conditions.

Projects with AI models and machine learning

Platforms, predictive models and AI tools that are built into real-life user and business scenarios.

Universal AI platform AI2Media

Universal AI platform AI2Media

AI2Media is an innovative cloud platform for automating the collection, analysis, generation and promotion of media content using artificial intelligence. Integration with WordPress, Bitrix, GPT, API for export, network of bots for auto-exporting.

Learn more →
STIPS Oculus — Trading & Analytics Platform

STIPS Oculus — Trading & Analytics Platform

Trading and analytical platform for cryptocurrencies: forecasts, signals and trade automation.

Learn more →
AI analyzer of Yandex and 2GIS cards for lead growth

AI analyzer of Yandex and 2GIS cards for lead growth

A web application that parses company cards from Yandex.Maps and 2GIS, analyzes them using AI and produces a list of specific recommendations for improving visibility and conversion to applications.

Learn more →
AI agent for realtor of the agency “Everything for Rublyovka”

AI agent for realtor of the agency “Everything for Rublyovka”

AI assistant for Rublyovka realtors: generation of news, descriptions of objects, selections for clients, analysis of areas and creation of images for publications.

Learn more →

AI Model Development Calculator

Preliminary estimate of budget and deadline. We determine the final estimate after checking the task and data.

Total

Description:Text classification and analysis, Cloud

Task for the AI model

Project stage

Data Status

Placement outline

Additional work

Total

Description:Text classification and analysis, Cloud
Project team

Who develops the AI model

We include only roles that are responsible for setting the problem, ML development and implementation of the solution.

AI / PRODUCT
Vadim Yurievich
Vadim Yurievich
Head of AI projects
AI / ML
Makar Kucherenko
Makar Kucherenko
Python/ML Developer
INTEGRATION
Victoria Smirnova
Victoria Smirnova
AI integration developer
You can assemble exactly the team needed for your project into a project.

What do our clients say about us?

Nbm-it on the map of St. Petersburg - Yandex Maps
December 19, 2023
Evgeniy I.
We have been working with NBM IT for the second year now; they have created an online store for us, support the website and help with improvements. Everything is high quality and prompt, special thanks to the manager Dmitry for his attention to detail!

Frequently asked questions about developing AI models

Your own model is justified if ready-made services do not provide the required accuracy, data cannot be transferred externally, or unique logic is required. For typical generation and recognition, it is often faster and cheaper to integrate a ready-made model.
Yes. First, we check the volume, quality, rights of use and absence of leaks of the target information. Then we create a dataset, a control sample and quality metrics.
There is no single minimum volume: it depends on the task, data variability and the underlying model. During the audit, we evaluate whether the current data is sufficient, whether markup is needed, or whether fine-tuning and synthetic examples can be used.
The preliminary cost of a PoC usually starts from 320,000 rubles. The production solution is evaluated separately and depends on data preparation, accuracy requirements, load, integrations and layout.
PoC usually takes from 4 weeks. MVP and production require more time: integration, load tests, security, monitoring and documentation are added to the training.
PoC answers the question of whether the desired quality can be obtained from the available data. The production model must additionally be load-bearing, error-handling, logged, securely updated, and have a rollback plan.
Yes. We support on-premise, cloud and hybrid hosting. The choice depends on data sensitivity, workload, available hardware, and response speed requirements.
The rights and composition of the transferred materials are fixed in the contract. For individual development, we transfer the agreed source codes, documentation and artifacts of the model; third party models are used under their licenses.

Leave your contacts - we will call you back, sort out the problem and offer the best way. We have more than 350 projects behind us, each of which we launched with an individual approach. We guarantee expert advice during business hours.

Contact information and details

We are for open communication.

Our team consists of experienced and proven employees. And therefore NBM-IT can guarantee you a high-quality result.

Saint Petersburg

m. Chkalovskaya,
st. Professora Kachalova, 11B

+7 (911) 814-86-78
on the map

Tampa, FL, US

12125,Lexington Park Drive, Tampa, FL

813-291-56-77
Antony

nbm-it@mail.ru

Services

Development

Corporate websites

Development on 1C-Bitrix

WordPress development

Online stores

Web Applications

Backend and highload

Services

Priority directions

CRM development

SEO promotion

Implementation of AI

RAG and knowledge bases

Touch panel software

Touch panel rental

About us

About us

Blog

Solutions catalog

Website Tools

Processing Policy
personal data

Partnership

Portfolio

Contacts