← Journal · · Note
Platformization: why it exists and whether you need it
Hey everyone!
Right now companies are actively adopting platformization as a tool that lets them cut time to market for new features as the product scales. Last year I checked out a meetup at Avito on this exact topic, and realized that folks on the Russian market are genuinely at the cutting edge of adopting platforms in development. That, by the way, is one of the reasons I moved there.
A matrix split into horizontals and verticals, picking the "best bike" and generalizing it to the rest of the teams, many companies go through these processes at different stages of their maturity right now. The Avito team released a short video explaining the basics of platformization, its benefits, and the history of how it was rolled out at the company. It'll be useful for anyone looking for arguments on why platformization is a good thing at large product scale.
Link to Avito's post itself.
As for ML, I usually split platforms into three layers: Data / ML / Inference, and all three actively interact with each other. The most detailed description of an MLOps platform I've seen is in this article, and its great breakdown in Russian here.
Platformization helps cut time to market, especially at large company scale: it reduces duplicate solutions, evens out quality, simplifies onboarding, and gives teams "rails" they can use to ship features to production faster. At the same time, platformization requires huge resources, a dedicated team of specialists, and infrastructure. So do you actually need it right now? I'll leave that question open for now :)