Serhii Borovykov · Product designer and AI builder

Each clock was synced by hand: a special crew drove around the rally, switched the clocks on and connected a computer to each one to set its time. If a clock switched off during the rally, there was no quick way to switch it back on with the right time. So clocks often fell out of step, and that caused a lot of problems with the results.

My idea was to take the time from GPS satellites. That way, it no longer mattered if a clock switched off: as soon as it was back on, it took the exact time from the satellites, the same time every other clock showed, wherever it stood.

It was a long black box on a tripod, with big red seven-segment digits and fluorescent lime-yellow caps at each end. The caps shaded the digits from the sun, and closed over the display like lids when the clocks traveled. I built them and wrote their software. They stood at the start and finish lines and at every time control, where crews get their time cards stamped.

That clock led me into timing rally competitions. There I built new timing equipment to replace the old: it recorded each car’s start and finish and printed the times on the spot. That printout was the official proof of who won. Then I took on another problem: safety. A rally spreads across a whole region, so after a crash it was hard to find out where it happened. I designed and built GPS trackers for the cars, and software that showed every car on a map at headquarters. From then on, if a car crashed, we could see where right away, and that made the rally much safer. Years later, that tracking software became my startup, Uboro.

A rally is a small society for a weekend. It takes a lot of money and hundreds of people, and the government too, because we close public roads. It takes months of planning. Then it goes live, and it can’t stop until it ends. Cars race through the stages, marshals stamp time cards, headquarters tracks every car, and the service park fixes them. On race day, the whole rally ran on the clocks, the timing and the tracking I built. One mistake can break all of it, and everyone feels that, at every level.

The peak was the Yalta Rally in 2012, a round of the Intercontinental Rally Challenge, shown on Eurosport. Two days, factory teams, about a thousand people organizing it, a budget of around ten million dollars, and observers from the FIA, the sport’s world governing body, there to keep it at the series’ level. At noon on the first day, all 35 clocks across the rally froze at 00:59:59. They had lost the satellites, and I couldn’t find out why. For the next hour I felt every bit of the pressure, and the guilt: the clocks were mine, and ten more minutes of it would have canceled the rally. At one o’clock they came back on their own.

That night I searched for the cause and couldn’t find it. We knew it would happen again at noon, so for the second day we issued a bulletin, the official notice to every crew: a pause from noon to one. At noon the clocks froze again, right on time, and the rally waited. Back home, with fresh eyes, I opened one clock and found the bug at once: one comma in the wrong place. The clocks ran on a microchip I programmed in assembly language, where you write every instruction for the chip by hand, down to single bits, so one misplaced comma was enough to stop everything.

I lived through many days like that, ten to twenty-five rallies a year, and year after year, pressure stopped overwhelming me.

After the Yalta Rally in 2012 and the Aleksandrov Rally in 2013, colleagues from across Europe noticed my rally tracking and wanted to use it at the European Rally Championship. The final tests were planned for 2014. That year Russia occupied Crimea and invaded Ukraine’s eastern regions. Racing in Ukraine stopped, and my rally tracking had no rallies left. We left behind everything that only a rally needed, kept what a fleet could use, and built on top of it. You paid by the day, only for what was being tracked.

As a founder I ran sales, marketing and the team, and I kept building: I designed Uboro’s screens and wrote their HTML and CSS myself. That’s where I became a product designer.

Smartcat had 11 standalone use cases, each with its own interface, across a three-sided marketplace of 500,000 freelance linguists, built for Fortune 500 companies. The experience sat on rigid legacy architecture.

We worked one case at a time, each with a before and after.

Activation rose from 7% to 85% and task completion from 55% to 98%. We launched self-subscription, and its conversion went from 0% to 5%.

I built the design team at Smartcat and ran it through both UX transformations. These are the principles and habits we ran it on.

First, I upgraded our project templates into agents, and a chat became a way in. Then agents that bring in other agents. Last, agents you give a goal, like a project manager that chases freelance linguists until every asset is done.

Engineers built the product while the company looked for market fit. The data model was heavy, and every engineer had changed the shared components in their own way.

I questioned the basics before designing anything: an object model and a flow model first, screens after. My first project was the feed, where every buying signal a Bridge finds lands: builders and viewers got their own modes, one feed replaced two sharing concepts, and it went live in about a month. The second was how a Bridge is made: plain words in, criteria and scores you can read out, the first feature I designed in the prototype environment.

Along the way I built Starbridge’s first design system, and the product started to look like one product. Then I hired two designers, the best I have managed.

With Claude Code I moved the design system into code, so design owns the components, and built a prototype environment on the same system. The first feature took 3 weeks of design work instead of 9 or 10, and real prototypes changed the discussion around it.

Specs in Linear are written from the prototype now, 5 to 10 times more complete than by hand. The whole design team moved over in 3 to 5 months, and product managers prototype their ideas there too.

I wanted the whole customer base heard. My team and I had proved the idea by hand at Smartcat, on 400 calls from one quarter. The tools I tried didn’t hold up, so I built my own there.

At Starbridge I built it again, starting with customer success calls. It finds praise and ROI for the team and for marketing, and gathers problems into a backlog that can prove or disprove what we are told. Every night, two AI passes read each new conversation: one finds candidate signals, the other throws out about one in four. Cards and a morning brief land in Slack, a web app holds the evidence, and anyone can ask Claude.

Once that pipeline worked, other teams asked for their own. New customers were sometimes oversold or confused, often about AI credits. I added lanes for onboarding and for sales trials, reading calls and trial emails.

Now the team that owns onboarding fixes the bottlenecks its signals show, and sales leadership coaches reps from what happens in trials. Customer success and sales use the signals for their accounts, the wider company reaches them mostly through Claude Cowork, and the backlog feeds the roadmap with real data.

About 210 pages lived on Webflow, and an outside agency built every new one.

I rebuilt the whole site in five days with Claude Code, on templates and the same design system approach. Then I set up a Slack channel with Claude where marketers ask for changes in plain words.

Six marketers onboarded. Changes went from weeks to dozens a day, and pages are now made for each ad cohort.

Marketing needs product images and flows to tell the story.

The prototype is a full copy of the product, so I made it the source. Flow tests in Storybook became HTML graphics. Anyone can ask for one in Slack, and agents watch the product and update graphics everywhere.

It took a few skills and a few tools stitched together, because the rest was already built.

Sales, business development reps, ads and outbound all make their own versions of marketing’s templates. Once everyone had AI, those versions drifted further from the brand.

I built an asset library where product graphics and Slack requests land, and the marketing team approves what goes out. Then I grew the website design system into one for all of marketing, with slides and social posts, connected to the library, Claude Cowork and Claude Code.

What the company makes now comes out on brand, with product graphics as current as the product.