Public Speaking

I love getting on stage to share lessons from building production AI systems and open-source machine learning tools. These are my talks over the years, from AI agents and evals to bringing machine learning to the Ruby and Go communities.

AI Engineer World’s Fair: Evals Are Broken, Use Them Anyway (2026)

My most recent talk argues that benchmark numbers deserve skepticism, and so does judging models on “vibes” alone. I walk through a three-zone failure-analysis framework for pulling concrete improvements out of broken benchmarks. Cline climbed from 43% to 57% on Terminal Bench through harness fixes and prompt engineering, all while keeping the same model.

AI Engineer (Europe): Don’t Build Slop, 4 Levels of AI Agent Maturity (2026)

A talk on why most teams over-engineer their agents. I lay out four levels of agent maturity and five rules for writing your own agent code, argue that the best agents are simple state machines with pruned prompts, and make the case for Kanban boards as the right UX for managing many parallel, inference-bound agents.

EuRuKo Ruby Conf: Introducing Tensorflow Ruby API (2017)

At Europe’s largest Ruby conference in Budapest, I introduced tensorflow.rb, the Ruby API for TensorFlow that I built with support from Somatic.io and the SciRuby foundation. The talk covers the core TensorFlow classes, image recognition with Inception v3, and visualization with TensorBoard, all from Ruby.

• Watch on RubyEvents

More Talks & Workshops

Over the years I’ve spoken at conferences across Europe and North America on machine learning, algorithm visualization, and life as a digital nomad in tech:

• DevOpsDays Recife (2022): Digital nomading in tech, talk
• GopherCon India (2019): Algorithm visualizations in Go, project
• Google Developers Windsor (2019): TensorFlow Ruby, event
• GoWayFest Belarus (2018): Visualizations in Go, project
• Ruby on Ice, Germany (2018): TensorFlow for Rubyists ML workshop, talk

If you’d like me to speak at your event, reach out on X (@arafatkatze).