MCP server connecting AI tools (Claude, Cursor) to Austria's open government data. Semantic search across 60,000+ datasets on data.gv.at in German and English. 18 tools for search, discovery, analysis, and data preview; quality scoring and schema inspection. Install with: uvx datagvat-mcp init. MIT, cross-platform.
AI-driven web application that combines generative design with real-world product data to create personalized interior concepts. Describe a room in natural language, generate an interactive 360° panorama, and discover matching IKEA furniture via Qdrant vector search. 1st Place – Bundeswettbewerb für Künstliche Intelligenz (BWKI) 2025.
Synth UI is an AI-powered SaaS platform that turns natural language descriptions into production-ready React components. Users describe the component they need in plain English (e.g. “a pricing table with three tiers and a CTA button”), and the system generates code that follows real-world patterns from a curated knowledge base of popular libraries such as shadcn/ui, Radix UI, and Tailwind-based designs. Under the hood, a modular agent architecture coordinates prompt engineering, semantic retrieval over a vector index of component examples, and structured output so that generated code is consistent and dependency-aware. The app offers real-time previews in the browser, support for multiple frameworks and styling approaches, and one-click export of deployment-ready code. Synth UI was built as a team project for the Bundeswettbewerb für Künstliche Intelligenz (BWKI) 2024, where it placed 3rd. The stack includes Next.js, the Vercel AI SDK, and a custom retrieval pipeline that balances similarity search with rule-based selection so that outputs stay close to established design systems and best practices.
Agentic system built at Hack-Nation Global AI Hackathon (Vienna/weXelerate) with Abdul Basit Banbhan. Analyzes 987 facilities in Ghana to detect medical deserts and connect patients with care. Five agents: Database (SQL + pgvector), Geospatial (road-network drive times), Medical Reasoning, Web Research, Mission Planner. Next.js 16, Vercel AI SDK v6, PostgreSQL + pgvector, deck.gl. Led our track in 36 hours.
I participate in competitive programming contests to sharpen my problem-solving skills. Here are my recent competition results.