Connect with us

Dmitry Petrov

Exclusive Interview With Dmitry Petrov, Co-Founder, And CEO, Iterative



Exclusive Interview With Dmitry Petrov, Co-Founder, And CEO, Iterative

As the machine learning market catches up with the competition, the ML engineers would need tools that can evolve beyond catering to the basic needs of an ML team, to make it easier and faster to develop models and enable collaboration. Iterative develops open-source tools for developers to build and deploy models to specialized software that can speed up the training process. Analytics Insight has engaged in an exclusive interview with Dmitry Petrov, Co-founder, and CEO of Iterative.

1. Kindly brief us about the company, its specialization, and the services that your company offers.

Iterative’s mission is to deliver the best developer experience for machine learning teams by creating an ecosystem of open, modular ML tools. Our tools are Git-native to bridge the gap between software engineering and machine learning so that these two sides of the ML to production pipeline can happen collaboratively, efficiently, and reproducibly.

2. Please brief us about the products/services/solutions you provide to your customers and how do they get value out of it

We build open-source tools that help ML engineers and data scientists automate their workflow and be more productive. DVC – Data Version Control – helps manage data and ML models and meld these two together with ML experiment tracking experience. ML engineers use DVC together with the Git source code version control system. CML – Continuous ML – is the way to use GPUs from clouds. It helps with ML-specific scenarios like using cheap spot instances for ML training and to recover jobs if they are terminated.


We believe in and support the open-source movement by contributing and maintaining modular open-source CLI tools for the greater ML/AI Community at whatever facet of the space, stage of the journey, or maturity level in which they find themselves.  Our SaaS product Iterative Studio, brings these open source tools together in a web UI that becomes the control center to drive collaboration, transparency, and efficiency among team members and leadership.

3. With what mission and objectives, the company was set up? In short, tell us about your journey since the inception of the company?

Iterative’s mission is to deliver the best developer experience for machine learning teams by creating an ecosystem of open, modular ML tools.  This mission was born out of the need for harmoniously joining the best practices of software engineering with machine learning projects that are driving a growing number of software applications today.  DVC (Data Version Control), the first tool of the suite, extends the versioning power of Git to the large and often unstructured datasets used in ML projects.  For the first time data, pipelines and experiments could be versioned reliably and reproducibly, supporting the data-centric movement and the growing need for data governance in ML projects. 

CML (Continuous Machine Learning), the next tool in Iterative’s lineup, provides CI/CD capability to ML projects.  This helps teams automate and orchestrate resources, whether in the cloud or on-prem.  The addition of TPI (Terraform Provider – Iterative) extends this functionality further, enabling the management of spot instances in multiple clouds to conserve resources when running large models.

MLEM, our latest tool, simplifies ML model deployment and is the basis of a full model registry, something many companies are working on in their ML teams.

We know that the one size fits all monolith platform will not work for all ML/AI teams. ML teams need flexible tooling that enables them to manifest their visions for their ML projects without the heavy lift of building that tooling from scratch. We aim to create these tools for them to allow them to iterate fast, use compute and people resources efficiently, enable cross-team collaboration, and provide as frictionless as possible deployment tools so that they can focus on solving their domain problems. We are building the tools to accomplish this task.

4. What is your biggest USP that differentiates the company from competitors?

Many competitors are building SaaS solutions that might satisfy the needs of data analysts and some data scientists, but ML engineers and researchers need an environment that is closer to a software development environment. ML engineers need command line, Git, CI/CD, IDE such as VScode, etc. This happens because ML engineers spend the majority of their time coding and we meet them where they are.

Some of our tools are GitOps-based, matching with software teams using Git as the source of truth for all applications. Information and versioning around ML models and experiments, data, artifacts, and hyperparameters, are all stored within an organization’s Git service. This GitOps approach aligns the ML model development lifecycle with that of an organization’s apps and services so they get faster time-to-market with transparent collaboration between software development and ML teams.

Healthcare is a huge driver of AI/ML these days. Healthcare companies, as well as hospitals, are working on computer vision-based solutions to better treat diseases. AI significantly increases the level of automation in manufacturing as well as logistics. Computer vision is becoming an essential part of inspection and quality control. The financial industry, fintech, and banking were always big consumers of AI and this trend won’t disappear.

6. What are Iterative’s company and business roadmap like for 2022?

We’ll be continuing to develop new tools for ML engineers and data scientists, with a couple of new ones being released later this year. On the commercial side, we’ll also be continuing to add enterprise-grade features to meet the demands of AI- and ML-first organizations that use our tools. These will include additional security controls as well as features around data discovery/management and further development on our existing tool set. We’re the first to offer a data catalog around unstructured data for machine learning use cases and will continue to innovate around this as the AI and ML market matures in the next 5 years.

7. What are some of the challenges faced by Data Scientists today with respect to machine learning and modeling?

As the machine learning and modeling space are in a nascent stage, there aren’t established best practices or MLOps tech stacks for data scientists and organizations in general to emulate. A lot of the tools on the market today aren’t built with data scientists in mind – they focus more on business users. With this approach, the developer experience isn’t the best, and models under development are hampered and have a slower time to market.

8. What are your predictions regarding ML and MLOps over the next 12 months?

We predict a shift to a developer-first experience in ML. More integration of ML tools to development environments such as IDE, version control as well as GitOps-based ML model deployment systems. ML engineers will embrace more and more software development best practices. Data-centric AI is a big trend in ML model development. We expect that some of these solutions will be democratized and released as widely available open-source tools.

The post Exclusive Interview with Dmitry Petrov, Co-founder, and CEO, Iterative appeared first on .



Get our daily News updatesSignup to get instant updates straight to your email