T-Bank Web: From a Monolith to Independent Blocks

Viktor was born in Kamchatka; at five the family moved to Nizhny Novgorod, which he counts as home. After school came a technical college, a former radio-engineering school: coursework on designing boards and routing tracks, and an internship laying fibre between buildings in Moscow's Tushino district, sitting in attics waiting for an admin to restart a router. A friend showed him how to build web pages while they played Counter-Strike, a demo of about three minutes; Viktor says it changed his life, and Alexander notes that such networking is usually underrated: you want people nearby whose craft still looks like magic. Then came his own sites, SEO and affiliate schemes, work resold to search-optimisation people, and a city forum whose engine he brought back on PHP in a day: friends spread the links, users arrived, there was almost no money in it, but experience accumulated.

That was enough for a first job won in fifteen minutes: after an interview at mail.ru's Nizhny Novgorod office he was called with an offer on his way home. He did not stay long — at nineteen he had to choose between full-time study and earning, and the study lasted two months. Next came a three-person startup — himself, a technical director and a business stakeholder — launching a news portal, worked round the clock and at weekends, entirely by choice. Then a large web studio turning out near-identical sites on a single CMS, including one for a big car dealership, where the working day became predictably dull. He moved to TKS Bank in April 2012 because he wanted to live in Moscow: the technical director then agreed the offer in a twenty-minute phone call, and the culture shock was the absence of tight control — one colleague even freelanced during working hours. Six months in the business stakeholder called him a valuable employee, and that kept him there.

For the first three years Viktor owned acquisition: landing pages, ad campaigns, full production access and direct conversations with business stakeholders, and at meetings his part was skipped — Firsov always had things in order; he describes that role as lead and senior developer at once, and calls the senior label a stretchy notion. He met React on a New Year landing page where Father Frost tipped presents out of a sack for the user to catch. At the same time the neighbouring team, building the personal account for business clients, picked Angular; why the paths diverged Viktor does not remember. Alexander retells a legend from people involved: the candidate pipeline was too thin, and a second framework let them take those who had passed the technical sections but did not want to write React, while splitting the hiring flows too. Viktor allows the premises may have been those, but history is silent.

The business wanted a marketplace with no separate personal account: an authenticated user roams the whole site and orders products, and there was even an attempt to broker other companies' products, mortgages he thinks, which never came together. Rewriting the application processes took about half a year, the stakeholder took the wait badly, and out came a monolith joining the public site and the personal account: the wait for a release grew to twenty or thirty days, could not be forecast, and it needed long, painful stabilisation before every release. An admin panel arrived, but it lacked flexibility for content, though routes and page parameters could be configured; outages and long downtime came round fairly regularly — the first sign the new stack migration had gone wrong. Alexander adds that frontend developers had designed it for themselves, hence the workarounds: a careless rollout on the frontend means an F5 for every user at once. In parallel they worked on a contract with a personalisation system for segment landing pages and search traffic, but tailoring pages to segments never happened, and two years later it all had to be redone.

The monolith was forked at the load balancer, with the application-process URLs cut off by prefix. The team had grown from two roughly peer developers to five or seven people and was verticalised around products: no shared backlog, no functional silo, with leads Viktor and Alexander placed into the new teams. Delivery got faster, but cross-cutting functions suffered: the redesign Viktor had promised took a month and a half to roll out because neighbouring teams kept breaking his tests. Alexander states the rule: team autonomy is paid for with the loss of the horizontal layer, and it has to be restored through platform machinery, not coordination by projects. Hence Viktor's idea: the blocks of the public site had to become independent of the application, published as npm packages or loaded at runtime. Micro-frontends were not in use then, and who brought them in first is still argued about; today teams far beyond acquisition use the page constructor, and an investors' site was assembled in two weeks.

In 2022 the company had to drop foreign infrastructure abruptly and weigh the risk of losing its mobile apps: the business focus had been on mobile since roughly 2018, the web held roughly fifteen per cent of the app's functionality, and about forty epics were counted as critical gaps. The personal account ran on virtual machines in an internal data centre and would not have survived a jump in load; the pitch to the business was not making the web great again but the third leg of the stool, without which it falls over. By the time the apps were pulled from the Apple Store and Google Play the architecture was worked out: the new main screen was started a week before the events of late February 2023, finished on 29 April, stabilised for a fortnight and shipped during a subsystem outage caused by the old account. That year the team grew from about fifty to a hundred and seventy people, roughly seventy developers were hired from the market, the main screen, QR payments and transfers by phone number went live; early the following year the web and mobile interfaces were unified through WebView in three weeks.