Budapest, Budapest, Hungary
Every product I build starts with the same question: why are decisions still made on data nobody trusts? Co-Founder at Logicl Lab — we build AI reasoning infrastructure for industries where legacy data, operational complexity, and a wrong answer cost real money. Maritime fleet operations, commodity origination, B2B industrial distribution. Deterministic, hallucination-free, deployed above the ERP companies already run. Measurable in 4 weeks. Before Logicl, I spent three years building production AI for financial markets at Collective Forecast — an environment that breaks every clean-room assumption. We designed a patent-pending architecture that hard-separates data computation from LLM reasoning, built measurable hallucination controls, and validated it against NIST AI RMF standards. That work became the technical foundation logicl runs on today. Always open to connect — CFO or CTO wrestling with legacy data, investors who think vertical AI is underrated, or founders building in hard industries.
We build AI reasoning infrastructure for industries where legacy data, operational complexity, and a wrong answer cost real money. Plugs into existing ERPs and legacy systems — no rip-and-replace, no hallucinations. Our data orchestration core handles dirty data: scanned PDFs, decade-old archives, mismatched naming conventions, mixed-language databases. My focus: go-to-market, product strategy, and enterprise partnerships across maritime fleet operations (LogiSea), commodity origination (AgriCore.Brain), and B2B industrial distribution. Think of it as giving a company's existing systems the ability to actually reason about their own data — and act on it in seconds.
Built and shipped a production AI analytics platform for financial markets — an environment that stress-tests every assumption about data quality, latency, and reliability. The core technical outcomes that now underpin logicl's platform: → Designed a Router-Engine-Agent architecture that hard-separates deterministic data computation from LLM reasoning — the foundation of our zero-hallucination approach across all products → Built CoFo Scout: a patent-pending multi-branch neural network that processes noisy, real-time market data across four simultaneous timeframes (1h / 4h / 1d / 1w) and outputs auditable, sourced signals → Developed and implemented a QA framework aligned with NIST AI RMF 1.0 and ISO/IEC TR 24028 — targeting hallucination rate < 5%, with SOC 2 readiness built in from day one → Shipped a multi-agent orchestration system (World AI) validated on real-time consumer data with a hard-validation layer achieving 99.8% factual accuracy — proving the architecture generalises beyond a single domain
Ukraine's largest road and infrastructure construction holding. First direct experience digitizing operations at enterprise scale — legacy databases, fragmented processes, and change management across a large organization. → Designed and implemented the company's BPM framework from scratch, supervising full rollout across departments → Built a staff performance evaluation system that delivered $300K+ in annual savings This role shaped a core conviction: enterprise AI fails not because of the model, but because the operational data underneath it was never built to be machine-readable.