Shipping, not just Talking
Real products solving real problems across healthcare, marketing, agentic engineering, AI research, and social impact. Each project taught something that theory alone never could.
bencium-harness
A build loop that no vendor can switch off. Provider-independent guardrails and workflows that build your software, run your research, and write your content — living in your repo as plain commands, not inside any one company's product. Switch tools, models, or vendors and the same six-step loop keeps running.
hypeless.io
Four drafts. Zero hype. Per post. Drop a URL, describe an idea — get four ready-to-post drafts in different angles.
insurgent-campaign
A grassroots-first campaign Claude/Codex plugin for anyone being outspent. It refuses the establishment playbook — content calendar, ad spend, influencer outreach — and treats earned trust as the real bottleneck, paid media as amplification on top of content that already works. Ships as a Claude Code plugin, portable to any agent harness.
brandguide/AI
The world's first brand building agentic harness platform that helps anyone build their brand through guided conversation. Your personalised AI mentor.
MedHubAI
Medically controlled AI conversational platform deployed across 10 clinics in production. Fully compliant with healthcare regulations, handling real patient interactions daily while maintaining complete auditability.
Anna, the Oncopsychology Assistant
Peer-reviewed AI assistant providing 24/7 emotional support to cancer patients and their families across Hungary since 2023. Developed with oncopsychologist Dr. Ágnes Riskó and published in Magyar Onkológia (PubMed PMID: 41819064) — validating that human-centred AI can complement clinical psychosocial care at scale.
Bencium Skill Marketplace
Open-source marketplace of agent skills compatible with 40+ AI tools. Works with Claude Code, Claude.ai desktop and browser, Claude Cowork, OpenAI Codex, and more. Install production-ready skills for design, architecture, code review, and deployment directly into your AI workflow.
AGI Detector
Early warning system monitoring leading Western and Chinese AI sources for real AGI signals, with cross-referencing and severity classification (low, critical, etc.). Uses historical trend tracking and pattern recognition to distinguish normal AI progress from potential AGI breakthroughs, based on signals like recursive self-improvement, meta-learning, or cross-domain generalization.
Regional Automation Processes
Dashboard measuring automation vulnerability of regional labour markets, combining employment data, land registry data, and automation methodology with AI analysis. Monthly-updated early warning system broken down by cities, where job automation and real estate market instability could trigger dangerous processes.
Agentic for Everyone
I am building a browser-based agentic AI interface bringing the power of developer CLI systems (Claude Code, Codex, Gemini CLI) to non-technical users. Streaming responses, tool visualisation, subagent delegation—without terminal syntax or cryptic errors. The paradigm: there is no interface between you and the work.