Gino Tesei
Most intelligent systems look impressive in a demo. The real challenge is making them reliable enough to matter. I explore that transition through evaluation, portable architecture, applied machine learning, agentic systems, open source, and research, grounded in experience building search and trading platforms and advising technical organizations across Europe and the United States.
Designing Portable Agents
Provider-Neutral Agent Systems in TypeScript with Open Standards
October 2026publication
Models and providers now change faster than the systems built around them. A better model appears, a provider changes its terms, or a workflow needs a different capability, and switching should not require rebuilding the application or surrendering control over data, tools, and evidence.
Designing Portable Agents shows how to change models and providers efficiently without pretending they are equivalent. It develops a provider-neutral TypeScript architecture based on model adapters, typed tools, reviewed skills, explicit trust boundaries, capability evidence, task-specific fallbacks, durable state, observability, and production gates.
An end-to-end research agent demonstrates the architecture across documents, images, spreadsheets, public sources, isolated computation, multiple providers, and a verified presentation. The result is not autonomy at any cost, but a system that can evolve with the model landscape while keeping authority, provenance, execution, and operational risk under control.
The current edition is a technical preview. Implementation and production-readiness claims remain evidence-scoped; compatibility does not imply full conformance.
Read the bookResearch & speaking
View all research- Learning End-to-End Patient Representations through Self-Supervised Covariate Balancing for Causal Treatment Effect EstimationA self-supervised representation-learning method for causal treatment-effect estimation, published in the Journal of Biomedical Informatics.
- Generating High-Fidelity Privacy-Conscious Synthetic Patient Data for Causal Effect Estimation with Multiple TreatmentsResearch on generating realistic, privacy-conscious synthetic patient data with known causal effects across multiple treatments.
- Fairness in Synthetic DataA panel discussion on synthetic data’s potential to address healthcare disparities, the distinction between associational and causal fairness, and why fairness-aware methods require sustained investment.