Sergej – AWS, Typescript, Node.js

A self-hosted infrastructure setup running over 30 Docker services across multiple servers, managed with GitOps. It covers deployment, secrets, backups, reverse proxying, and observability, with new servers provisioned automatically.

• managing 30+ Docker services across multiple servers with Komodo GitOps (stack definitions synced from Git, automated pull-and-deploy);
• managing secrets with Infisical, using pre/post-deploy hooks for automated secret injection;
• running a Traefik reverse proxy, Restic backups, and a Grafana/OpenTelemetry/Loki/Tempo observability stack;
• provisioning new servers end-to-end with Ansible.

A Solana token trading platform with automated buy/sell strategies, leaderboards, referrals, and a lottery system. Available as a web app, a browser extension, and a public API. At peak: ~11,000 on-chain transactions per day executing in 600–700 ms, ~300 daily active users, and ~$300k in Solana traded in a single day at ~99% uptime.

• built the full stack solo: NestJS microservices backend, SvelteKit web app, and a Svelte browser extension in an NX monorepo;
• built a gRPC service for real-time blockchain data ingestion with local caching and instant position updates, enabling sub-second auto-buy/sell execution;
• set up per-PR preview environments with isolated databases, Redis, and full app stack behind auto-generated domains, with migrations run automatically;
• optimized CI/CD to ~5 minutes from push to a running AWS ECS task, using parallel builds across self-hosted runners and nx affected to rebuild only what changed;
• built end-to-end type safety across backend, frontend, and extension via shared Valibot/Drizzle/proto schema packages;
• set up Infisical-based secrets management and a full observability stack (Uptime Kuma, Grafana/OpenTelemetry/Loki/Tempo/Mimir);
• led 2 developers to ship a new feature into production.

An AI-driven job-matching platform that maps a person's skills to over 3,000 ESCO occupations across more than 2.3 million European job postings, using semantic search over multilingual data.

• built a semantic search using a fine-tuned multilingual E5 model and a Weaviate vector database;
• processed multilingual datasets, using GPT-3.5 for automated label correction and fastText for language detection;
• built a skill-relation recommender system;
• built a community forum with Svelte and Strapi.

An internal employee marketplace web application enabling colleagues to browse, post, and manage listings within the organization.

• built the marketplace full-stack with Spring Boot and PostgreSQL on the backend, and Svelte with Tailwind CSS on the frontend;
• implemented core marketplace features including listing creation, browsing, and user management.

A real-time NFT sniper bot for a pay-to-play blockchain game on Binance Smart Chain. The bot evaluated whether a champion NFT was worth buying and executed purchases instantly by competing against other bots in BSC's pending-transaction mempool. Earned ~$300/day at peak.

• designed a real-time valuation formula to determine whether a target NFT would be profitable to buy;
• built the mempool interception logic to front-run competing bots through transaction fee optimization on BSC;
• operated the bot to ~$300/day in peak earnings.