Ilya Mokhov

May 2025 · as a product designer

Another VPN that never launched

By the time it was ready, the market was already significantly overheated

There were no extra chairs, so the project is going into the drawer forever

Some principles is —

external consistency with other saves cognitive resources;

upsell mechanics appear in the right context;

aha-moment when getting acquainted with the app

In the early stages, we totally borrowed the animation from a site with fancy backgrounds. If the app hadn’t been ditched, we would’ve made our own since we knew how to do it

Maybe another time

Ilya Mokhov

the card promo as another card in the stack

May 2024 · as a product designer

Nefin is a crypto wallet that also does fiat

We grew Nefin out of a cash exchange counter in Kazakhstan

My manifesto for the project was to cut off everything unnecessary and put the app back together out of the pieces it could not work without. I built the information architecture on that. Three screens, and any action sits two taps in

Other finance apps copy each other. I took the heaviest scenarios apart and rebuilt them. The transfer came out as one straight line

I talk to the user in text, the one language they already read. I kept icons only where they carry one meaning. The fewer shapes I left, the faster the user reads

When the user sends money, the screen already holds everything the decision needs. The form they fill in is the line they read back, so I built no review step

Topping up, swapping, sending, signing up: four different mechanics on one architecture. The user meets it on the sign-up screen

send: the amount and the address as type

the invoice created, ready to send as a link

Invoicing is what separates this wallet from other crypto apps. I built it on the same frame as everything before it

I wanted the user to finish what they came for and leave. I was not trying to keep them here

The repetition made the build cheap. We assembled new screens out of parts that already existed instead of drawing them, and a bank this size took far less design and code than its size suggests

the admin cockpit: money at risk, revenue and antifraud

June 2026 · as a product designer

Bot protection for ads traffic

Agencies buy ads and bots eat the budget

Clyro catches them and collects the proof to claim the money back

Live prototype

the client account: traffic, bots and protection

I wanted an MVP built out of the simplest parts I could find — so the loop was PRD > Claude Code > Figma

two tables, the same cell holding different value types

This is an admin panel, so almost everything on screen is data —
I made the table cell first and gave it 27 types of value

the cell taken apart into its value types

After that everything turned out to be the same cell in a different wrapper: the bot sources widget, the dropdown form for a new user, the contact block you edit in place
I reused it 31 times across the app

a block takes one to four modules of the grid

A card stacks up, a table keeps its width and scrolls

One screen, the same pair three times over

May 2024 · as a product designer

Blask counts who is winning the gambling market, country by country

Three designers, more than a hundred and fifty Figma files. I show four big jobs I did

blask.com

the new navigation in four states at once

I built the header and the filters with Claude and handed them over as working code

In the old navigation tabs sat inside tabs. Going from a brand in India to the same brand in Brazil meant walking back up through them

I pulled the context into one line at the left edge: the world or a country, and the slice of data next to it. Inside a brand card the same line switches its slices and takes you back up

A task on its way to the developers

the affiliate badge spec on its way to the developers

the old widget header against the new one

The old widget header took three rows before the chart started. The new one is one line at the same left edge: title, metrics, periods

It holds a chart, a table, a widget with no chart at all. The team stopped drawing headers and started picking a state of this one

the analyst drags widgets into their own layout

The analyst who watches one country and the one who watches five brands no longer get the same page. They build it out of the same widgets

One payment form with seven states

The plan is a table of 126 countries against the modules, and the price adds up as you tick the cells. The checkout panel stays open next to it, so the total moves while you tick

I assembled the plan out of parts the product already had: tables, filters, tier badges, chips. The customer works with the product before they pay for it

May 2026 · my own project

The Watcher checks what a trading channel actually did

A channel tells its readers what to buy and when. The calls that worked stay up, the ones that missed get edited, half of them are spoken in a voice note or drawn on a screenshot. Nobody checks six months of that by hand

The Watcher checks. It listens to the channel around the clock, turns every post, voice note and chart into a record of what was promised, then looks up what the price actually did

A month of a channel is just a list of calls, so the screen is a feed. One line per call: the price it started at, the price it ended at, what the market did in between. Colour only on the direction and the money

the order line taken apart into its states and controls

A review agent walks the live app and checks every new screen against this one

Not every message is a trade. Someone predicts, someone admits what they already hold, someone claims they won, someone is paid to post. Counting a prediction as a trade is how a channel's score turns flattering

Every call gets two endings: the one the trader claims, and the one you get if you cap what a losing call may take. The grey third number is that second ending, and the distance between the two is the whole product

I drew that line first and built the rest of the app out of it

Then I tested the obvious idea: what if you simply repeat the calls. Over a falling month, 52 finished calls lost 4.8% each on average. But 14 of them, closed by my own cap, later reached the trader's target. The damage came from that cap, and the real difference was between channels, not between rules

The Watcher turned out to be not a machine that finds good trades but a way to tell a pattern that works from one that only looks like it does. I pointed it at my own work first

An idea gets replayed on old data, with the system seeing only what was known back then. Five of mine went through that, and all five came out negative