From Physics to Engineering: Curiosity Pays Off Later

Andrey was born in 1989 in Vladimir and grew up in Raduzhny, a small town in the Vladimir region closed in Soviet times and code-named Vladimir-30 after its distance from the regional capital. Laser weapons were built there: the cold war had reached the point where everyone could make missiles, and the idea appeared of putting lasers into orbit to shoot them down. Alexander recalls Reagan's space programme; the guest carefully replies that he would not judge whether this was an answer to it. A side effect of the closed town was a concentration of technically educated people: after perestroika some left to do science in Moscow and America, but many stayed, and two strong schools grew up there. As Andrey remembers it, in ninety-six a fellow townsman took gold at the international physics olympiad. He studied in lyceum classes and at the correspondence physics school of Phystech, where assignments arrived by post and were marked by Phystech students; at sixteen he left for the specialised science and education centre at Moscow State University and from there entered the radio engineering and cybernetics faculty, drawn partly by its Intel department.

At Phystech he met classmates who were already writing 3D engines with physically honest rendering, and realised that science holds people far more talented than him. Alexander adds an anecdote he thinks came from a book about Jeff Bezos: he brought a problem to a brilliant fellow student, who thought for some fifteen seconds and produced the answer without writing anything down, after which Bezos moved into engineering, where there is no winner-takes-all rule and second-tier researchers do not stay unknown. Andrey describes the same feeling through processors: at a seminar you are still working out how to solve the problem while someone has already answered, as if your architecture were a generation old. It is a humbling experience, yet by his account one of the best periods of his life. The turning point came in the second year, when the informatics teachers proposed doing coursework in teams: Andrey's group wrote a Skype killer with many-to-many conferencing, around 2007 or 2008, when neither Google Meet nor Zoom existed.

His first attempts were in gamedev, during the boom of applications inside social networks. Andrey launched his own startup when VKontakte let application developers work with third-party advertisers: single-handedly he built a platform for monetising their traffic, with development, product, marketing and design on him, a LAMP stack and a PHP banner rotator. Developers came, saw that there was no traffic, and left; the plan was to run between them fast enough for a critical mass to gather, and once the servers even stopped coping. Encouraged, he wrote to Oleg Tinkov on Twitter: would you like a banner in a place visited by 300 unique visitors? No answer came, and the person responsible for customer acquisition at the bank wrote a month later, when the project had already been abandoned for his diploma. Then came Clickberry, with an interactive layer on top of video, and after it OneTwoTrip: two million lines of untyped Node.js, air tickets, and the launch of car rental, where after the product manager left Andrey also became the product lead of a team of two and a half people.

He took that product to its first sale, heard "show us the money and we will give you developers", returned to engineering and soon became a team lead. For a month and a half, in headphones, he optimised the heaviest step of search with its hundreds of pricing rules: timings improved by around thirty per cent, he says, without remembering the exact numbers, the virtual machine count dropped by tens of per cent, and a beautiful step appeared on the Grafana dashboards — only to creep back release by release, because the monolith's shared code ate the gain. That new love of microservices got him scouted by Tinkoff, which wanted to decompose its monolithic CMS: it ran the site and the landing pages through which the bank acquired millions of customers, had been built by frontenders for frontenders, and fell over entirely if someone misconfigured a cache. Alexander recalls his own document on separating services by criticality to shrink the blast radius.

The Rancher orchestrator dragged the team into the office at night more than once, and together with Stas, who owned infrastructure, they moved to Kubernetes and Google Cloud — the first team in the company, as he recalls; the standard parts of the services were packaged by Dima's Crab framework, which reduced building a microservice to assembling ready-made pieces. After the big release Alexander promoted Andrey to a manager of managers with four teams, but the role was short-lived: the wish to work in an international company became burning, and out of two offers he chose the management track at a British startup building a customer intelligence platform — analysis of reviews, surveys, tickets, emails and call-centre calls. His six years there began with a team of six, serious downtimes and infrastructure debt; the Google Cloud migration paid off again, except that this time the infrastructure had to be staffed by hiring.

Managing in another language proved a separate challenge: directness normal in Russian-speaking teams reads as tactless in the West, and his focus on hard skills shocked candidates who expected a chat about their CV, though it let him hire strong people. The product rested on aspect-based sentiment analysis: a set of topics was devised for the client's industry, a sample labelled, a network trained and its accuracy measured. Then LLMs arrived — a powerful model became available through an API, the barrier to entry collapsed, competitors sprang up like mushrooms and clients promised to repeat it over a weekend at a hackathon; engineering came to matter more than research, though, Alexander objects, in B2B people buy a solution rather than a technology. Andrey left for scale — a global B2C mobility service grown out of a taxi app, where price balances riders against drivers. On his time budget he warns that the checksum will not add up: roughly 12–15 per cent each on people and on running projects, about 30 per cent on technical work and 15 per cent on shaping future ideas.