How it all began
Every year, thousands of ML models are created at Sber. And all would be well, but many teams regularly spent time developing solutions that already existed. The same tasks were being solved over and over again, losing valuable resources. The reason was simple: there was no single platform where you could quickly find, test, and use already available ML models.
What I was about to do
The team was tasked with creating a user-friendly marketplace platform, similar to Hugging Face, where data scientists and developers can easily find the models they need, test them right on the spot, and integrate them into their products via an API.
I understood that the project would not be easy, but it is precisely such challenges that I like.
What exactly did I do
At the start of the project, the analyst and I conducted a series of interviews with 30 users: data scientists, developers, and validators. This helped me clearly understand their daily tasks, pain points, and expectations for the future platform. I also studied the experiences of leading market players such as Hugging Face and Kaggle to understand how to make the product truly user-friendly.
I created a detailed customer journey map (CJM) and based on it, developed an interactive prototype, which we gradually brought to an MVP. Among my proposals was the implementation of a simple ability to launch and test models directly in the platform interface, as well as a highly transparent display of key performance metrics for each model. This significantly reduced the time from finding a model to its launch in production.
When we showed the alpha version to users, I actively collected and analyzed their feedback. It was this that helped refine the product and make the interface intuitive and truly useful.
What results did we get?
The coolest result is that the time spent searching for the right model has decreased by 90%)). In just six months, the number of published models has increased by a third, and according to survey results, the platform has proven to be convenient and clear for 85% of our users.
Perhaps the best indicator of a project's success is that users themselves have actively begun to recommend the platform to their colleagues.
What important thing did I take away for myself?
In this project, I once again realized how important feedback and real empathy for users are. The coolest solutions emerged not when we just followed the plan, but when we listened carefully to people and adapted the product to their real needs.
I also learned to work better with metrics and clearly demonstrate the effectiveness of my decisions to the team and clients.
It turned out great — the platform is already bringing real benefits, and I am proud that my solutions have helped users to make their work easier and faster.
The tools and methods that I used
Pixso, Board, ClickStream, CJM, User Flow, user interviews, and usability testing