Portfolio
A focused overview of products and systems I designed and built.
Filter by type
Web solutions
Corporate website for IC-Aumont — a licensed staffing agency operating in Slovakia, focused on industrial hiring and long-term workforce solutions. The platform is built as a high-conversion marketing tool: clear service structure, fast navigation, and direct paths for both candidates and business clients. Job listings are easy to browse, while contact flows are optimized to reduce friction and increase inbound leads. The design follows a strong, recognizable visual identity (red/black/white), with bold typography and clear CTAs. Mobile-first approach ensures smooth experience across devices, especially for job seekers. SEO-ready architecture with clean URLs and structured pages improves visibility in regional search, while accessibility and performance optimizations guarantee fast loading and usability.
A dark, minimalist personal brand site for Alex DDK: bold hero typography, project highlights, a short philosophy section, and clear contact paths. Editorial content is managed through Notion as the backend—copy and project entries can be updated in Notion and surfaced on the landing page without redeploying the whole site. The layout stays intentionally quiet: high contrast (black, white, and accent gold), generous spacing, and responsive imagery so the first impression reads as a curated portfolio rather than a generic template.
SaaS systems
Attendance Analyzer is a SaaS-style internal tool built for workforce operations, designed to process messy real-world attendance data and turn it into clear, actionable insights. The system analyzes uploaded Dochadzka .txt files, handling inconsistent formatting, encoding issues, and duplicate records without breaking. Instead of relying on file structure, it validates each worker against actual calendar days, ensuring accurate detection of critical issues such as unreported absences and missing check-ins. A clean dashboard provides instant visibility into key metrics, with severity-based highlighting to quickly identify problems. Coordinators can review records, copy relevant data, and export results to CSV for further processing. Built with React, Vite, and Tailwind on the frontend, and powered by a Node.js + Express backend, the application is optimized for performance and reliability. It supports multilingual use (Slovak / Russian / English), light and dark themes, and can be distributed as a standalone desktop-like tool via bundled deployment. Designed for real operational environments, the system prioritizes reliability over assumptions — making it a dependable tool for teams managing high-volume workforce data.
A full-stack SaaS authentication and authorization system designed for secure, scalable user management across modern applications. The platform supports multiple authentication flows, including traditional email/password and OAuth 2.0 (Google, GitHub), combined with optional two-factor authentication (TOTP) for enhanced security. Session handling is implemented via JWT stored in HTTP-only cookies, ensuring both security and seamless user experience. Role-based access control (RBAC) enables fine-grained permission management, allowing applications to define roles and control access at scale. The system is built with a modular architecture, making it easy to extend, integrate, or adapt to different product requirements. On the backend, Node.js with Express and TypeScript provides a reliable API layer, while Prisma ORM ensures flexible database management (SQLite by default, with easy migration to PostgreSQL). The frontend is powered by React and Vite, delivering a fast and responsive interface. Designed with production use in mind, the system balances developer experience with real-world security needs, making it suitable as a foundation for SaaS products, internal tools, or multi-tenant platforms.
Public marketing site for DirectHeader: fast landing with hero, product sections, social proof, and clear CTAs toward signup and demo. Implemented as a static-first front end with optimized assets, responsive layout, and accessible typography and focus states. Content is structured for SEO—semantic headings, meta tags, and stable URLs; contact and lead forms post to a small API with spam protection and optional CRM/email hooks. Performance is tuned with image formats, lazy loading, and minimal layout shift so Core Web Vitals stay in the green on mobile. The design system reuses tokens for spacing and color so future pages (blog, changelog, legal) stay consistent. Screenshots and product imagery can be swapped in as they are produced—placeholders are in place until final art is ready.
Morettinero E-Commerce
AI automation
An internal automation layer that routes incoming support tickets and operational requests through LLM-assisted classification, entity extraction, and policy checks before they hit human queues. The goal is not to replace agents—it is to remove repetitive triage, reduce misrouting, and surface structured summaries so the right team sees the right ticket faster. The pipeline ingests email and form payloads, normalizes them into a canonical schema, and runs a retrieval-augmented step over internal knowledge (FAQs, runbooks, product boundaries) to propose category, priority, and suggested next actions. Confidence scores and guardrails determine when a draft reply can be suggested versus when the case must escalate unchanged. All model outputs are logged with trace IDs for audit and regression testing when prompts or retrieval corpora change. Downstream, the same orchestration hooks into webhooks and CRM updates: auto-tagging, SLA timers, and handoff notes for specialists. Built with API-first services, queue workers for burst traffic, and human-in-the-loop review for sensitive categories—typical of production AI automation rather than a one-off chat demo.
Backend
A reference implementation of a distributed platform using domain-oriented services, asynchronous events, and a service mesh for traffic policy. Synchronous calls stay within bounded contexts; cross-domain effects publish to a durable event bus so consumers can react without tight coupling. Kubernetes hosts the workloads with health probes, resource requests, and pod disruption budgets tuned per service. gRPC is used for internal APIs where strong contracts matter; REST remains at the edge for public consumers. Message brokers (e.g. RabbitMQ-style patterns) handle backpressure and retries with idempotent consumers to avoid duplicate side effects. Observability is unified: distributed tracing across the mesh, structured logging with trace IDs, and RED-style metrics per service. Security includes mTLS between mesh members, rotating credentials, and network policies that restrict lateral movement. The result is a system that can evolve team by team—deploy independently, scale where needed, and reason about failures without a single shared database becoming the bottleneck.
A production analytics workspace built for teams that need to see what is happening in their product right now—not yesterday’s batch job. The platform ingests high-volume event streams, normalizes them in near real time, and pushes live updates to the browser over WebSockets so dashboards refresh without manual refreshes or polling. On the backend, ingestion workers scale horizontally behind a partitioned queue; hot paths are optimized for predictable latency, and aggregation pipelines roll raw events into time-bucketed metrics stored in PostgreSQL with Redis as a fast cache for latest counters and leaderboards. The front end is a React application with charting tuned for large datasets: virtualized tables, debounced filters, and saved views so power users can pin the slices they care about. Operational tooling includes per-tenant rate visibility, dead-letter inspection for malformed payloads, and configurable retention so storage costs stay under control as volume grows. The system routinely processes well over a million events per day in production-like load tests, with monitoring and alerting wired into the same metrics the customers see—so regressions surface before users report them.