Building and Owning Specialized Enterprise AI with Umi

At a TechCrunch Disrupt session, Manolis, formerly of Google Cloud and now leading Umi, explains why enterprises should build and own specialized AI models instead of relying only on general-purpose frontier models. The presentation combines strategic arguments with a demo of Umi’s platform for automating AI engineering.

■ Strategic case for specialized AI
- IP ownership, data leakage, economics, sovereignty, privacy/security; examples include Revolut, Harvey, banks, and legal firms
- Framework for build vs. use general models, with programming agents as a common exception

■ Umi platform and FAI agent
- FAI automates evaluations, synthetic data, training, deployment, and monitoring; open-source package supports text/images, enterprise platform supports text
- Emphasizes ownership of weights, data, and recipes; one-click deployment, autoscaling, and production quality tracking

■ Q&A topics
- Context layers vs. model training, distillation, synthetic data, and gathering scattered enterprise data; ideal user is an organization that sees AI, not ML infrastructure, as its competitive advantage

Suited to enterprise AI leaders, CTOs, founders, and machine learning engineers evaluating build-versus-buy decisions. Viewers gain an overview of strategic criteria and a platform workflow for creating specialized AI, and can consider whether to start building their own models sooner rather than later. No pricing or Amazon product listings are presented.