TypeScript · React · Next.js · Node.js · Python · AI integrations 📍 Prague, Czech Republic · Remote (CET) · 🟢 Open to roles & contracts

I build products end-to-end — from requirements and architecture through to production deployment. Frontend on React and Next.js, backend on Node.js or Python, AI integrations on top of LLM APIs, deployment on Vercel and Railway.

Most of my commercial work sits in three areas: Telegram automation, AI-integrated dashboards, and data extraction platforms. I've also owned the non-code side — writing technical specs, scoping and pricing projects, and running client communication directly.

Frontend: React · Next.js · TypeScript · JavaScript · Vite · Tailwind · HTML5/CSS3 Backend: Python · Node.js · Express · FastAPI · Fastify · aiogram · grammY · REST · WebSocket · JWT Data & queues: PostgreSQL · pgvector · SQLite · Redis · BullMQ · ARQ · Prisma · Alembic AI: Anthropic Claude · OpenAI API · Google Gemini · structured outputs · RAG · MCP servers Tools & Deploy: Docker · Docker Compose · Vercel · Railway · GitHub Actions · Tailscale · Git

• Remote Jobs Hub — vacancy aggregator over Telegram channels. Six services in one compose file: an MTProto collector reading whitelisted sources only, a rule-based prefilter that drops noise before any LLM call, structured-output extraction into normalized fields, and two-layer dedup (exact + semantic via pgvector). Ships to users as a Telegram Mini App with HMAC-verified initData auth. Python · FastAPI · Telethon · ARQ · PostgreSQL/pgvector · Next.js

• LeadRadar — Threads lead pipeline: headless scraping, Gemini relevance scoring, Telegram delivery. Two BullMQ queues on Redis sit between the stages, so a Playwright crash doesn't lose scraped posts and a transient 429 doesn't kill the run. Scores are cached by sha256(post text) — the same post seen under several keywords costs a Redis GET, not a second billed call. TypeScript · BullMQ · Redis · Playwright · Gemini

• antiTCK — anonymous incident-report bot with mandatory human moderation. Every submission is sanitized before publishing — phone numbers, emails, @usernames, URLs, licence plates — and coordinates are rounded to district level. There is no auto-publish path in the codebase; that's an invariant, not a setting. Node.js 22 · TypeScript · grammY · Prisma · PostgreSQL · Vitest

• crypto-widget — resident desktop widget: a docked pill that expands into a live Binance watchlist with charts and spike alerts. Native window behaviour (always-on-top, docking) handled in Rust, because the web layer can't express it. Tauri 2 · React 19 · Rust

• TON Testnet Wallet — self-custodial TON wallet, Telegram-style UI, live updates over WebSocket. Shipped, then audited: 25 fixes across key handling, input validation and balance-refresh races, with the suite going 78 → 347 tests. TypeScript · React · Vite · WebSocket · Vitest

• Wallet Analytics Dashboard — transactions, token flows and balance history for any Ethereum address on the Etherscan V2 API. A later audit closed a publicly reachable withdraw endpoint and took tests from 0 to 175. Next.js · TypeScript · Vitest

• Data Room MVP — due-diligence document room: nested folders, PDF upload and preview, instant search, IndexedDB persistence, no backend. React 19 · TypeScript · Zustand · Vitest

• AI Integrator — Landing — bilingual marketing site with 10 live Gemini-backed tools, two-tier Upstash rate limiting and validated email capture. Tools that need scraping unavailable from serverless are labelled as demos rather than faked. Next.js 15 · TypeScript · Tailwind · Framer Motion

• Agent Farm — hub-and-spoke agent platform: Telegram gateway, Fastify orchestrator, and per-VPS RAG workers. Workspace isolation is enforced by Postgres RLS (SET LOCAL app.workspace_id), not by application code; the only public surface is one webhook, everything else listens on a Tailscale mesh, and the model never gets a shell tool. TypeScript · Fastify · BullMQ · PostgreSQL RLS · Docker · Tailscale

• Chronicles — a text game where the dice are rolled by code, never by the model: the LLM names a skill and difficulty before the roll and narrates a result it didn't decide. The deterministic engine has zero dependencies and zero API calls, so balance is debugged for free — 500 stubbed runs caught four defects, including one where succeeding at checks made a skilled player die sooner than a random one. TypeScript · Anthropic API · Vitest

More in the repository list: an AI IDE with a RAG pipeline and NATS-orchestrated agents, an Instagram content-intelligence bot (Whisper + OCR + LLM function calling), and several production Telegram bots.

Team Lead / Full-Stack Developer · Freelance · 2023–2024 Led 4 developers on commercial Telegram bots — code review, architecture decisions, client communication. Shipped production bots with auth, payments and admin panels.

Full-Stack Developer · Freelance · 2022–Present Web apps, dashboards and automation for paying clients. End-to-end delivery from spec to deploy.

I use Claude and Copilot for boilerplate and for moving through unfamiliar APIs. Architecture, review and debugging stay with me. Pick any repository here and I'll walk you through why it's built that way and what I'd change — that's the part worth judging.

• 📧 aleksfialko15@gmail.com
• 💼 LinkedIn: Alex Fialko
• 🌐 Portfolio: alexfialko.vercel.app
• 🕐 CET · 🗣 EN fluent · RU/UA native · CZ basic

Open to remote Full-Stack / AI engineering roles and contract projects.