Aleksei Nizhgorodov

Frontend & architecture, led the team· 2026

Designed and built a production VPN cabinet: OAuth/OTP login, one-link activation across 6 platforms, subscription migration from a Telegram bot, and a custom censorship-resilience layer. White-labeled into 8 brands from one codebase.

• A commercial VPN needed a multi-brand web cabinet, so I built the whole SPA: file-based routing with auto code-splitting, a root store of 8 domain stores, and a hand-rolled API client that Zod-validates every response. One codebase compiles into white-label builds for 8 brands via a build-time flag.
• Under blocking, the cabinet has to stay reachable even when the main domain is killed. I built a failover layer — a service worker pulls a list of live mirrors and redirects to the first reachable one, with a static fallback page, and the mirror-list endpoint is hidden behind obfuscation. The cabinet still opens when the main domain is blocked.
• Legacy subscriptions from the old Telegram bot had to move into new web accounts. I wrote a 5-step migration with cross-account login and payment detection by watching the subscription term grow against a stored baseline — it works across accounts and handles the “already migrated” case.
• Firebase, Lottie and QR had bloated the bundle. I added route auto code-splitting, manual vendor chunking and lazy-loaded the heavy pages — the initial bundle dropped from 1706 to 656 KB, gzip from 467 to 175 KB.

Frontend: real-time, payments, chat · team project· 2025 — 2026

A team product where I owned the WebSocket generation engine, the payments layer, and the legacy-chat integration. Midjourney, GPT-Image and LLM chat behind one interface — with real-time and subscriptions.

• Midjourney generation inside Telegram had to show live progress, so I designed a custom WebSocket client (EventEmitter + MobX) with a request registry: every action gets a UUID and an entry with its own state machine, a 7-minute per-request timeout, and auto-reconnect. The channel survives connection drops.
• I wanted the full Midjourney toolset, not just generate, so I built a generation-action engine on MobX that models the history tree with parent-linked nodes — a complete set of operations (vary, upscale, zoom, pan, blend, reroll) on top of an unstable backend.
• I owned monetization end to end: tokens, plans, subscriptions — purchases, pay-in-external-browser, promo codes, balance polling. Real payments in production.
• The legacy chat had to fold under one roof and ship, so I set up the deploy (Docker + nginx) and embedded a separate Next.js chat app via iframe with a Telegram↔iframe bridge — safe-area and keyboard over postMessage, haptics relayed through. One deploy, seamless chat inside the hub.

Frontend · map layer & rendering

Several large Leaflet projects: geozones, real-time markers and long tracks with timelines — with custom layer logic and rendering optimized for thousands of objects.

• The map had to hold thousands of markers and long tracks without lag, so I wrote my own layer logic and tuned the render — clustering, thinning, drawing on Canvas and WebGL. It stays smooth on large datasets.
• For real-time monitoring I added live marker updates, geozones, and tracks with timelines — an operator can scrub through movement history by time.
• The map had to be reused across several products, so I pulled it into a set of components with one API — a shared cartographic base for several team projects.

Frontend, BFF & data layer· 2026

In ~2 weeks: a backend-for-frontend, a resilient ingest pipeline and ~20 clinical dashboards over medical terminals — a custom Russia map, read-layer RBAC on medical data, and blood-pressure classification per the national guideline in SQL.

• A B2B medical-analytics product had to stand up from scratch in ~2 weeks, so I built a pnpm monorepo: a Vite + React + MobX frontend, an Express BFF, and a shared package of Zod schemas as the single source of truth. The browser never sees the upstream credentials, and types run end to end from DB to UI.
• The medical-terminal API was unreliable and I couldn’t lose records, so I wrote my own client and an idempotent ingest: backoff over 6 attempts, a circuit breaker (8-failure threshold, 60s cooldown), 7-day windows with cursor pagination, and an incremental cron. Backfill and sync survive outages with no manual restarts.
• Medical data needs more than hiding it in the UI, so I split it into separate tables and put RBAC at the read layer — a narrow role is never JOINed to the medical fields. Sensitive data is unreachable on the backend, not just hidden on the client.
• There was no off-the-shelf geo engine for a raster Russia silhouette, so I wrote my own lat/lon → image-percent projection (two-corner calibration, antimeridian normalization) and a map component with proportional bubbles and drill-down — an interactive choropleth right in the dashboard.

Frontend · architecture from scratch, team reviews

Frontend architecture from scratch for a surveillance and analytics platform on TS / React / MobX — architecture, code review, team mentoring.

• I built the frontend of a video-analytics platform from scratch — designed the architecture on TypeScript / React / MobX, set the patterns, ran reviews and mentored the team. The team keeps building features on that base.
• I built real-time dashboards on Recharts — operators see analytics live, without reloads.
• Large lists were sluggish, so I virtualized them, memoized the heavy work, profiled in Chrome DevTools and cleared the memory leaks. The UI stays responsive on large datasets.

Frontend · library author

A custom report library from React components: renders WebGL maps and charts into PDF, with auto page numbering and headers/footers. jsPDF / react-pdf couldn’t do it — so I built my own.

• We needed reports with maps and charts, but jsPDF and react-pdf can’t render WebGL maps or complex layout — so I wrote a library that assembles a PDF straight from React components and embeds the WebGL maps and charts. The reports look like the product UI, with no hand-built PDF layout.
• For multi-page reports I added automatic page numbering and running headers and footers — print-ready documents with correct pagination.
• To avoid duplicating layout, I made a single render layer: the same React components drive both screen and PDF — one source of truth for the UI and the reports.

Frontend · two clients + native on Capacitor, led the native team

A dashboard on a large touch screen + a parent PWA: two clients sharing state. Led the team building native apps on Capacitor (iOS / Android / Android TV).

• The product lived on two very different clients — a gamified dashboard on a big touch screen and a PWA for parents. I designed both around shared state with real-time sync, so the screen and the parent’s phone always show the same thing.
• It had to ship to the stores and Android TV, so I led the team packaging the web app into native builds on Capacitor (iOS, Android, Android TV) — one codebase reaching several platforms.
• I added a Service Worker for offline and auto-update — the app works without a network and updates itself.

Frontend · MVP architect· 2026

In ~4 days, built a window tablet that reacts to the queue over WebSocket (no relay backend), entertains and shows contextual ads during service, then collects the official service-quality survey.

• Each tablet had to figure out on its own when a ticket was called to its window, with no relay backend — so I wrote a self-reconnecting WebSocket client that diffs queue snapshots and derives call / recall / finish events by its own window id. Per-window routing with no backend of its own.
• There was a race between a fallback timer and the operator’s real “finish” event, so I wrote a state machine idle → called → serving → rating → thanks with stale-event guards and killed a duplicate-survey bug. The flow runs clean, with no double survey.
• I wanted to fill the wait at the window with relevant ads and entertain slower visitors, so I built a “service → life-situation” heuristic, legally-compliant ad formats, and a mini-game that tunes its difficulty to the player’s reaction time — contextual ads plus a game even older visitors can play.
• Animations on the tablet stuttered, so I moved the slide progress bar from a JS width animation to CSS keyframes scaleX — compositor-only, no layout. The bar runs smoothly, on the GPU.