AI that works in the real world.

AI Research & Engineering leader Founder · exited Conference speaker

I'm Viktor Tisza. For 15+ years I've turned machine learning into products and businesses, and built the teams that ship them.

Work with me See projects

Startup exitAcquired by a US firm

Krónikás · open source

3continents forecasting elections

15⁺years in machine learning, data science and AI

1AI startup co-founded, built and acquired

3continents where my election model is in use

AI Leader Research & Engineering

Co-founder & Chief AI Officer OneAdvisor.ai, acquired

AI Advisory Board Eulytix

Speaker Compass AI & Tech Summit 2026

Open-source author Krónikás

Alumnus Corvinus University of Budapest

About

Scientist, engineer and founder.

The hard part of AI is rarely the model. It's the evaluation, the data, the people, and the decision the model is meant to change.

Today I lead AI Research and AI Engineering, taking generative AI from experiments to products customers depend on. Before that I spent years where models meet money: quantitative research, insurance, retail and enterprise software.

I co-founded an AI company and saw it through to acquisition. I advise startups, speak at conferences, and still write code. Some of it now forecasts elections around the world.

Expertise

Depth across the whole stack.

From Bayesian models to LLM agents to the operating model that gets them shipped.

Generative AI in production

LLM agents, retrieval and tool use, built with the evaluation pipelines and observability that make them safe to put in front of customers.

AgentsRAGEvalsObservabilityMCPTool use

Exit.

Co-founder and Chief AI Officer of OneAdvisor.ai, acquired by a US firm.

ML & data science

Bayesian modelling, forecasting, deep learning and classical ML, chosen for the problem.

PyMCForecastingVision

Building AI teams

Hiring, growing and running research and engineering teams that ship together.

Org designHiringCoaching

AI strategy & advisory

Taking organisations from pilots to measurable impact. AI Advisory Board member at Eulytix.

RoadmapsGovernanceAdvisory

Projects

Side projects the world now uses.

Built in my own time to stay hands-on. Some grew far beyond that.

Open source · Python

Krónikás

Election forecasts that show the full range of outcomes.

A hierarchical Bayesian model that turns opinion polls into probabilities: each party's likely vote share, the plausible range, and the chance of finishing first. It learns every pollster's lean, and says plainly what polls can't tell you.

Built for Hungary, now forecasting elections around the world.

EuropeLatin AmericaMiddle East

kronikas.eu GitHub

Illustrative · synthetic polls

Live forecasts

SzázKilencvenKilenc

Data-driven Hungarian election forecasting.

Named after the 199 seats of Hungary's parliament. An independent forecast published in Hungarian and English through the 2026 election, with Krónikás running underneath.

szazkilencvenkilenc.hu

Game design · built with AI

Kétharmad

A satirical strategy card game.

Sixteen years of Hungarian politics in 162 cards and four factions. Designed with AI as a creative partner, then balanced by AI players over 300,000 simulated games.

Explore the game

Lab notebook Earlier experiments +

Chili ClassifierIdentifies 142 chili varieties from a photo, running in the browser. 2nd place at OTDK, Hungary's national student research conference.Open →
ElektoRSimulates how different electoral rules would have changed a Hungarian election.Code →
Game Title EvaluatorA recurrent network that predicts a game's review score from its title alone.Code →
Rental SpyderScraping and visual analytics of the Milan and Budapest housing markets.Code →

Speaking

On stage, what actually worked.

Keynotes, panels and workshops, in English and Hungarian.

● 2026 · Budapest

Compass AI & Tech Summit

Forecasting an election in the open: how 199 and Krónikás were built, what the model can and cannot see, and lessons from publishing live forecasts.

Topics I speak on

Taking GenAI to production 01
Evaluating LLM agents 02
Building & scaling AI teams 03
Bayesian forecasting & uncertainty 04
From AI startup to exit 05

Measure before you scale

An AI feature without an evaluation is an opinion. Evals and observability come first.

Say how sure you are

Every number deserves a range. Honest uncertainty earns more trust than false precision.

Ship, then learn

Short loops with real users, deep science underneath. The best teams do both.

Contact

Let's build something real.

Leadership roles, advisory work, speaking invitations, or a hard modelling problem. I read everything.

Email me LinkedIn GitHub

vtisza@gmail.com