Talks about what happens after the demo.

01

The Invisible 90%: Why AI projects fail after the demo

Demos have never been easier to build. Production has never been harder to reach. What the demo structurally hides, what the billion-dollar failures have in common with a rescue I lived through, and the four steps out.

Audience

Directors, heads of product and senior managers, cross-industry

Format

30 minutes, including Q&A

Given at

AIPMX 2026 · Product Management 4.0 Strategy Day 2026

02

Fine-tuning open LLMs past the frontier models

How a well-scoped fine-tune on open models can beat foundation models on a specific task, at a fraction of the cost and with full control of your data and infrastructure. Told through MemeGPT, trained on Imgflip's meme data on NVIDIA H100 and Intel Gaudi2.

Audience

Engineers, data scientists and technical leaders

Given at

Generative AI World Summit & MLOps World · LLMs in Production · Analytics Vidhya

03

Optimising multi-GPU training at scale

Where large-model training breaks down across many GPUs, and the optimisation work, with MIT/ISTA researchers, that fixes the bottlenecks.

Audience

ML engineers and infrastructure teams

Given at

MLOps Community Munich