Prague · MFF UK · Zoplio
Engineer and founder, moving into AI-safety research.
I build systems, and I study how the systems we use to oversee AI break. My current question: when we read a model’s chain of thought to catch deception, does that oversight survive outside English — and when it fails, is it the monitor or the model that fails? I’m finishing my Bachelor’s at the Faculty of Mathematics and Physics, Charles University, and starting a Master’s in Artificial Intelligence there this autumn.
On the side, I’m co-founder and CTO at Zoplio — an AI scheduling primitive — with Lukáš Hellesch. We are backed by BUDETO Studio.
Chain-of-thought monitoring — reading a reasoning model’s visible “thoughts” to catch misbehaviour before it lands — is one of the few levers AI safety currently has, and recent work shows it is fragile: outside English, monitors miss deceptive reasoning almost entirely. What nobody has established is why. If the monitor is at fault, the fix is a cheap translation step; if non-English reasoning genuinely carries less monitorable signal, that is a fundamental limit. Two opposite policy conclusions — and a question a careful experiment can decide.
A monitor-language × reasoning-language evaluation matrix on agentic deception transcripts with programmatic ground truth, plus the deciding arm: translate the chain of thought to English before monitoring, and measure how much detection recovers. Translations verified by native Czech speakers are the data layer most labs can’t produce in-house. Built on open-weight reasoning models with the Inspect framework; code, prompt packs, and all transcripts will be public.
The AI scheduling primitive: a hosted API and SDKs (Node, Python, Anthropic MCP) any application embeds to delegate meeting negotiation. I own the agent runtime — agent-to-agent negotiation protocol, confidence-scored memory across categories, LLM-driven conversation engine, WhatsApp Cloud integration, Google Calendar OAuth, Stripe usage-based billing.
Backed by BUDETO Studio (€50k pre-seed).
An AI company platform co-founded with Lukáš Hellesch — autonomous AI agent teams (CEO, CTO, CMO, CFO + specialists) that plan, ship, sell, and close. Red Bull Basement ’26 Czech Republic national top 10. Wound down in 2026 to focus on Zoplio and research. Watching our agents confidently misreport intermediate results is part of what pulled me toward oversight research.
Android application for advanced Bitcoin asset management: M-of-N multisig (BIP-48, BIP-67), coin control, Trezor hardware-wallet integration via Trezor Connect, PSBT distribution between cosigners (BIP-174). Backend implemented as Kotlin/Ktor microservices on PostgreSQL with Docker Compose; frontend in Jetpack Compose. Verified on Bitcoin testnet with a complete 2-of-3 multisig transaction.
Supervised by RNDr. Filip Zavoral, Ph.D.
Built progressively from the bottom up: console output and a bump-pointer allocator, then thread scheduling, interrupt handling, virtual memory with page-table protection, system calls, custom minimal libc, and a multi-process userspace running user-mode binaries in isolated address spaces with stack-overflow detection. Tested on the MSIM simulator with a CI suite covering every milestone.
Multi-colony foraging simulator demonstrating emergent intelligence from local rules: pheromone trails with diffusion and decay, a customisable per-colony genome, parallel multi-colony execution, and a sandbox editor for map design. Emergence from simple local rules is the thread that runs from this simulator to my interest in what large models do when nobody is watching.
I’m happy to talk about chain-of-thought monitorability and deception evals, AI-safety research collaboration, or anything to do with running an early-stage AI startup.
Email — hi@kubadvorak.com
GitHub — github.com/Majny
LinkedIn — jakubdvorak-ai