MLOps (Machine Learning Operations) Machine Learning Methodology

DevOps Methodology

Main article: DevOps Methodology

Machine Learning

Main article: Machine Learning

Chronicle

2025: AI Model Management Pipeline. How ModelOps approach is gaining popularity in Russian companies - TA opinions

In Russia, against the background of the active development of its own AI stack, a new critical discipline is being formed - the management of working machine learning models. This is no longer just about deployment, but about a full-fledged DevOps for artificial intelligence: monitoring, automation of retraining and ensuring their uninterrupted operation in production. This material will focus on the ModelOps methodology, which TAdviser dealt with in early October 2025.

ModelOps (Model Operations) is a methodology for managing the lifecycle of machine learning models and analytics, offering tools to optimize the efficiency of introducing AI models into production, this is a set of capabilities that are primarily aimed at managing the full lifecycle of all AI models and decision making.

𝓲

Фото: Freepik

Svyatoslav Smirnov, head of the K2 NeuroTech division, claims that the ModelOps approach is actively used in Russia, especially in large companies where work with AI has moved from single pilot projects to industrial operation of models. Initially, many started with a narrower practice - MLOps, which focuses on the technical side of deploying and monitoring one model, the expert continues.

According to him, in Russia, one of the leaders in the implementation of ModelOps are banks and fintech companies. In the financial sector, hundreds of ML models are used daily in scoring decisions, risk assessment, anti-fraud and customer outflow forecasting. To effectively manage these models, special ModelOps platforms are used, which provide centralized control over the versions of all models, their metadata and artifacts.

Ekaterina Torsukova, head of Data Science "DAR" (part of KORUS Consulting Group of Companies), called MLOps one of the indicators of maturity in the field of AI. The methodology allows you to calculate the effectiveness of AI models, reduce the number of unsuccessful implementations due to constant quality monitoring, and also send to the product environment exactly those models that have proven their effectiveness excluding subjective factors. ModelOps plays an important role in the transition from a reactive to proactive strategy for managing AI models, but this approach is not yet common due to the insufficient maturity of the entire AI sphere, Torsukova said.

Ainergy AI Infrastructure Director Konstantin Kudryashov disagrees with the opinion about the weak distribution of ModelOps in Russia. According to him, this methodology has already become the norm. First of all, it is used by banks and telecom, there are also projects in retail and industry.

At Sberbank"" this AI Cloud and, ML Space through them, hundreds of models from scoring to assistants spin in the sale. T-bank MTS AI In and built conveyors from fichestor to automatic departure and monitoring., "" and NLMKSeverstal"" Gazprom neft use this approach for predictive maintenance and quality control, - said Kudryashov.

Maxim Zakharenko, CEO of the Cloud Computing IT company, also knows the cases of using ModelOps in Russia: telecom (primarily for detecting anomalies), industry (for predictive analytics) and retail (recommendations, dynamic prices). Thus, the Auchan Retail Russia network has introduced a corporate ModelOps platform for working with Big Data.

According to Svyatoslav Smirnov from K2 NeuroTech, ModelOps is used in the Russian industry to optimize processes and predictive analysis in production, and in telecom - to predict the load on the network and predictive maintenance of equipment. Operators often use this approach when implementing solutions that automate the technical support of the user and the work of administrators.

The founder of the IT company Netrack Alexey Rubakov draws attention to the shortcomings of ModelOps. First of all, this is the complexity of implementation, requiring investment in infrastructure. The benefit is manifested when managing dozens and hundreds of models, and only if a data management culture is formed in the company. If a company is overly addicted to tools, there is a risk of excessive complexity, more time and money is spent on supporting the ModelOps platform than not on the models themselves, he argues. The pitfalls lie in regulatory and ethical issues. This problem is especially relevant for banks, medicine, the public sector. It is required not only to control the technical correctness of the models, but also their compliance with regulatory requirements, safety standards and ethical standards, Rubakov added.

2024: MLOps Global Instrument Market Size Reaches $1.58 Billion for the Year

At the end of 2024, costs in the global MLOps instrument market reached $1.58 billion. More than a third of this amount was provided by the North American region. Such data are provided in a Fortune Business Insights study, the results of which were published on April 7, 2025.

MLOps, or Machine Learning Operations, is a set of practices that automate and simplify the development, deployment and operation of machine learning. Companies can use MLOps to optimize and standardize processes throughout the machine learning lifecycle. These processes include model development, testing, integration, launch, and infrastructure management. The modern MLOps ecosystem brings together many tools - from cloud platforms to open solutions. Each tool performs certain functions in the process of creating and supporting machine learning systems.

The implementation of MLOps provides organizations with a number of significant benefits. First of all, this is a significant acceleration in the launch of models to market: automated processes shorten the path from idea to industrial implementation. In addition, MLOps frees specialists from routine tasks, which allows you to increase the efficiency of teams. Automatic testing and monitoring ensures the stability of the models in operation. MLOps practices improve troubleshooting and model management in production. In particular, software developers can track system performance and reproduce their behavior to solve problems.

But there are also certain deterrents, one of which is security-related issues. Organizations need to protect sensitive data, secure access to models and infrastructure, and compliance. In addition, it is important to have a structured process for validating, evaluating, and approving models before they are put into effect. This may include checking for fairness, bias, etc. Analysts stress that working with legacy libraries is one of the most common challenges facing businesses. According to IBM's report on the use of artificial intelligence, about one in five companies claim difficulties in securing data.

According to the deployment method, the market is divided into cloud, local and hybrid segments. Cloud platforms offer end-to-end solutions for the entire machine learning cycle. However, the hybrid segment dominates, with security and cost concerns prompting most companies to adopt a balanced approach that combines cloud and on-premises data centers. In terms of the use of MLOps, IT and telecommunications, healthcare, BFSI (banking, financial services and insurance), production, retail, etc. The first three areas provided a significant contribution in 2024: for example, the share of IT was 27.6%. From a geographical point of view, North America leads with costs of $0.57 billion. Globally, the following are named major players:

• DataRobot;
• Domino Data Lab;
• AWS;
• Microsoft;
• IBM;
• NPE;
• Allegro AI.;
• Mlflow;
• Google;
• Cloudera.

Fortune Business Insights analysts believe that in the future, the CAGR in the market under consideration will be 35.5%. As a result, by 2032, costs could increase to $19.55 billion.^[1]

Notes