Live demos · 6

Built. Shipped. Running.

Working systems embedded here — not screenshots or slide decks. Open one, poke around, then use the contact form if you need something similar.

CAPTCHA Lab

Three providers, real server-side verification — Turnstile, reCAPTCHA v2, hCaptcha.

URL → AI encyclopedia with synthesized article, topic extraction, and knowledge graph.

YouTube URL → article, sentiment, word clouds, chatbot. €1 Stripe unlock.

One-time checkout, recurring subscription, and self-serve cancellation via Stripe.

Run private LLM models locally on your own hardware — Ollama + Docker + Cloudflare Tunnel, with a live chat interface.

Live HTML signature preview + send to inbox via Resend. One form, one click.

Projects · 4

Each entry follows the same shape — problem, what was built, tools used, result.

Programming / Automation / AI

Problem: Students manually refresh Facebook groups and listing sites to find rooms — slow, noisy, easy to miss listings.
Built: Telegram bot with filter onboarding (price, region, room type, preferences) and push notifications on matches. AI extraction subsystem normalizes messy listings before filtering.
Tools: Python, Telegram Bot API, Supabase, AI extraction, filter engine.
Result: End-to-end pipeline from scraped listings to filtered Telegram notifications. Filters across 6+ attributes; live in private testing.

AI Solutions

Problem: Rental listings on Slovenian sites are unstructured prose — hard to filter, sort, or match against user preferences.
Built: Extraction workflow that turns free-text listings into structured JSON — price, location, availability date, room type, restrictions, contact details.
Tools: Python, self-hosted LLM, structured-output prompting, JSON schema validation, database integration.
Result: Extraction subsystem feeding the Mojcimer bot. Converts arbitrary listing text into clean records suitable for filtering.

Programming / NLP

Problem: Product review streams are too large for manual reading; teams want sentiment trends in near-real-time.
Built: Java pipeline processing streamed product reviews — sentiment scoring with execution comparisons across sequential, parallel, and distributed modes.
Tools: Java, Stanford CoreNLP, WebSocket, parallel processing, MPJ Express.
Result: Working pipeline + benchmark report comparing execution modes — informs architecture choice for production sentiment systems.

Infrastructure

Problem: SaaS lock-in and per-seat pricing get expensive fast; some projects can't legally send data to third parties.
Built: Personal server environment hosting dashboards, services, bots, and AI tools. Secure remote access without exposing ports.
Tools: Proxmox, Docker, Linux, Cloudflare Tunnels, Ollama, OpenWebUI.
Result: Production-grade self-hosted stack — template for client deployments where data sovereignty matters.