How to Build and Scale Ultra-Lean Gen AI Startups
summary
Asserts that multiplying human creativity by GenAI (an M² mental model) beats chasing giant training clusters; shows a lean stack instead: a toolkit that does radex quantis and prune, a Rust runtime for fast inference, and radex HUB for private model distribution. Announces a granted project with Nvidia and University of Naples to run ~4,000 GPUs (roughly 336 days 24/7) on a €50M-scale effort, and demos a Polygen-style fact‑checking app and an mlops evalop platform—early pilots work and revenue‑generating startups pass $1M.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:13Why this talk is short
- 2:14Lego duck experiment
- 9:02Decoupling AI
- 12:54Ultra-lean Gen AI startups
- 21:24Unique hard problem
- 25:41Private datasets and explainable AI
- 27:0650-million-euro AI project
- 31:00Radex toolkit demo
- 38:03Polygen fact-checking demo
- 45:16Q&A: Hard problems
- 47:44Q&A: Green zone
- 50:14Q&A: Persistence
- 53:05Closing remarks
related talks
20mTrue Exponential: Building a Generative AI Company
44mKeynote: Tracking the Pulse of Generative AI
44mKeynote: Tracking the Pulse of Generative AI
17mGenerative AI: Keeping it real
52mSimplifying the adoption of generative AI for enterprises
54mThe Future of AI: Building the Most Impactful Technology Together
24mScalable GenAI in Production
23mScaling GenAI Applications and Agentic AI for Business Competitive Differentiation