Yevhen Peshynskyi

A customer data platform for airlines, designed in code instead of Figma. The usual way — a Figma mock-up, a handoff, weeks of building, manual testing — is replaced by a pipeline: research, an HTML prototype in two variants, my pick, then planner, coder, verifier and reviewer agents in Claude Code build it on the design system, and a tester skill walks every flow as a real client, role by role. In the last comparison it caught 70% of what a human tester found, and the human missed 80% of what it found. I lead the front-end architecture in a team of five: the structure, the design system — 96 components in three tiers, with Storybook as the reference — and the logic of every feature. About half of the front end came out of my sessions.

One booking engine for any airline, built by AI agents — a harness of 5 agents and 32 skills — without Figma, by one person. Where a new airline normally means a month of restyling, here each airline picks its components in the admin and its whole style is one JSON — a day's work. It began with research into competitor airlines, platforms and UX patterns. The front end talks to any airline's backend through one contract: it asks which of 28 capabilities the backend can serve and hides what it can't. The next step is the main idea: a front end that adapts to each passenger, composed from Paximizer CDP advice — the order of steps, the offers, the layout — with the total price always in view.

A presentation and content studio that does the work of a marketing department: no designer drawing decks, no manager picking materials. Every week it researches what to post about and plans two weeks ahead, with briefs. Decks, posts, narration and video are built on one design system with templates and rules, voiced by a free local voice. A web admin runs it all — the calendar, decks, posts — and decks are generated on the server by an AI model; a Telegram bot handles approvals and urgent orders, minutes instead of days; published posts are cross-posted to paximizer.com.

I package a product completely — the interface, the site and all the sales and promo materials — so no agency or marketing design team is needed. Each site starts with research — competitor sites side by side, what they say and what they miss — then wireframes, the design and the promo kit. paximizer.com replaced Webflow and kept all 91 legacy URLs, so not a single link broke; wind2ride.app carries route share pages, club invites and a video guide. Many more sites since 2010 are offline now or under NDA — I show them on request.

I'm a cyclist, and I build an app for cyclists — alone, from the idea to the stores, doing what a small product team does: the product, the design, the build, the releases and the support. It began with a ride into a headwind and an analysis of how wind data can score a route, segment by segment. Riders tested it in a beta and reported from the app with screenshots; then came the social side — clubs and group rides. On Google Play, and on the App Store since summer 2026.

A knowledge base an AI can ask — a context generator. A question about the product that used to take hours of digging through old code and docs, or days of onboarding for a newcomer, is answered in seconds, from a business rule to a line of code. I documented the old front end and back end, indexed it into a RAG, and built an MCP server, so any AI agent could ask the project and get full context for any task.

A fully customizable booking engine for airlines — used by 20+ of them. Brand customization moved from manual restyling, weeks per airline, to design tokens applied in days. Research came first — airline constraints like PNR, SSR, fare rules and baggage — then the booking logic in wireframes, design tokens for multi-brand theming, the interface, tests on clickable Figma prototypes (+5% conversion in an A/B test, preliminary), and a handoff to the front-end team.

A customer data platform for airlines, designed in Figma and taken from MVP to production by one designer. The design rested on real API models from four airline systems, so the screens matched the data from the start and the front end was built once, without redesign loops. Most of the work was variants: every component drawn in several versions, tested against each other, the right one kept — across analytics, passenger profiles, loyalty, campaigns, support and monitoring, with roles, states and edge cases.

A constructor kit for airline websites — one template restyled for any airline (white label) — so a new airline's site is assembled in days instead of designed from scratch over weeks. Research into what every airline site needs, 20+ wireframe screens tested, semantic tokens with light and dark themes, a component library with variants — and a constructor: pick a variant of each block and the site assembles in the airline's brand.

I can automate any workflow, and every tool here removed a repeated manual step for good. When a step repeats, I first look for a tool that already does it, and build one when none fits: skills, plugins and apps for real, repeated work. Kryakl began as a tool for myself and grew into a free app my friends use. Muse, a Claude Code skill, designed this page the way a studio would, in one session: 23 directions, an independent jury, an idea board, 3 sketches.