15 Sep, 2026CERNA.AI Festival (Ostrava, CZ)event ↗

Evaly v praxi: od narativu ke ground truth

A 45-minute hands-on workshop with Veronika Kráľ. What LLM evals look like and what they are actually for, on one small real project: a bot that scores rental listings against a written brief. We walk through the three ways we got the grading wrong — one holistic narrative, then weighted criteria with rubrics, then ground truth for what the model can't read — and what each mistake taught us.

The worksheet we handed out — Flat Scout: pracovní list, the exercise that turns the two-page narrative brief into weighted criteria with rubrics — is a printable PDF ↗. The deck is downloadable too ↗.

2026AI jako zrcadlo, ne sluhaCERNA.AI Festival (Ostrava, CZ)

A 12-minute keynote on the CERNA AI stage, delivered as "Kognitivní výfuk: k čemu je AI, která smí jen číst?". Everyone is shipping agents that answer your inbox and order things for you. Is there still a point to an AI observer that only reads — email, calendar, journal, browser — and assembles what it finds in your digital exhaust? I argue yes: a read-only personal AI sees more, breaks less, and has a much easier conversation with GDPR, the AI Act, and your own paranoia.view ↗

Airplane mode for your second brainPKM × AI Meetup #1 (Prague, CZ)

An experiment report for the first PKM × AI meetup, a sibling of evals.cz that I co-organize with Jakub Žovák. Can a customized pi coding harness on a local model do what Claude Code does for my notes, offline, on a plane? What Claude Code gives the vault and what pi had to grow, honestly scored; which local models were tried on which questions; and what worked. The verdict: enough for bounded work, not yet for an always-on observer, because the model eats the machine.view ↗

RAG a evalyAI Monday Jihlava #7 (Jihlava, CZ)

"RAG is dead" is the take in every other thread in 2026 — and it's wrong: naive retrieval-augmented generation is still a sensible default, beaten only in some cases, and measurement is the only way to know if you're one of them. This talk walks the retrieval pipeline end to end, then turns to the part that matters — telling whether your RAG actually works, with ground truth, retrieval metrics, RAGAS, LLM-as-judge, and error analysis feeding an eval flywheel.view ↗

Choose Your Ground Truth: A Field Guide to Synthetic Data for EvalsEvals.cz Meetup #3 (Prague, CZ)

You need an eval set but don't have a hundred real production failures to build it from, so you reach for synthetic data — and most first attempts quietly produce garbage. A field guide to the techniques that actually work, from real-incident seeds to personas to RAG-grounded generation, with one throughline: synthetic data needs its own eval, so choose your technique backwards from the eval you want.view ↗

Want this talk for your audience?Invite me to speak ↗