Fullstack engineer specializing in AI-native systems, multi-agent workflows, and high-performance architectures. I build things that have to stay up: event-driven pipelines, low-latency data layers, developer tooling, and the occasional C++ simulation engine at midnight.

I like building systems where orchestration, state, and performance actually matter. Finance trackers, game engines, self-hosted infrastructure, weird automation projects — the domain changes, the fun part doesn't.

Matnabru is my personal scratchpad: the stuff that doesn't make it into a Jira ticket.

• Orange Polska // Fullstack Engineer, AI Engineering & Automation — Building web/mobile apps and microservices that integrate AI orchestration, STT/TTS pipelines, and enterprise APIs (Microsoft 365, OAuth 2.0) into production-grade internal tooling.
• Aexol // Software Engineer — Outsourced across multiple client projects (3–5 months each), building and expanding products across different domains and stacks. Highlights: eliminated Stripe rate-limit bottlenecks via event-driven sync (Webhooks + MongoDB), increasing read throughput 200×; contributed to bare-metal K8s migration (Hetzner), significantly reducing infrastructure OPEX.
• Deloitte // Salesforce Intern — Where the first commit happened (2021).

• Primary: TypeScript (Node.js/Next.js/NestJS), Python (AI/Automation/Lazy side projects), C++ (low-level, heavily optimized systems).
• Infrastructure: Kubernetes, Docker, GCP/AWS (Serverless & Event-driven), Linux Internals.
• Data & AI: Postgres/Mongo/Firebase for the truth, Redis/Valkey for the speed, RAG pipelines, Vector DBs, GraphQL/Swagger for the interface.

A Google ADK + A2A-powered LeetCode learning system that routes between cached knowledge, similarity search, explanation, and fresh multi-agent solving workflows.

• Architecture: Hybrid policy layer over a deterministic Parser → Architect → Coder → Librarian execution path.
• State: Persists solved problems, pattern notes, and similarity links into an Obsidian-backed knowledge base with ChromaDB.
• Tooling: Distributed A2A services, inspectable Agent Cards, ADK Web observability, and a clean dev workflow for tracing multi-agent runs.

A darkly comic factory builder / tower defense game about a burned-out grim reaper managing a bureaucratic afterlife skyscraper. Private repo — code walkthrough available on request.

• Architecture: Three-tier game stack — C++ GDExtension simulation layer → C# gameplay systems → Godot rendering and tools.
• Systems: Vertical floor-by-floor factory building, drafted soul-wave defense, workforce recruitment from defeated enemies, and persistent office infrastructure across failed runs.
• Tooling: In-engine worldbuilding and simulation tooling designed to author large, highly systemic game spaces from inside the game itself.

Data ingestion engine for unified portfolio state across fragmented financial markets. Private repo — code walkthrough available on request.

• Integrations: Interactive Brokers (Stocks/Options), PKO BP (Open Banking/PSD2), Binance/Bybit (WebSockets).
• Challenge: Merging high-frequency WebSocket streams with asynchronous, rate-limited Banking APIs into a single low-latency query layer.
• Stack: Node.js, TypeScript, Redis, MongoDB.

Built in a day for live football watch parties with friends. Scrapes today's fixtures via proxy rotation, pulls live event odds from a dedicated API, and tracks user predictions and scores in MongoDB.

mateuszurbanek.com • LinkedIn