TypeSafe AI

TypeSafe AI develops AI models for software automation, focusing on affordable intelligence and reliable execution. The lab operates in stealth and offers a public waitlist. Its System1 model is designed for machine-to-machine execution, targeting compliance pipelines, real-time decisions and autonomous agents. The intended approach embeds reliability into the model itself. The company also describes TypeSafe Language Models as programmable, modular models designed to compose with software.

Founded in 2024, TypeSafe’s founders are CEO Diogo Almeida, CTO Erik Gafni and COO Sasha Sheng. Almeida previously helped develop reinforcement learning with human feedback at OpenAI; Gafni brings experience building production AI systems, while Sheng worked in research engineering at Meta/FAIR. Almeida’s technical writing examines how models learn: his analysis of distillation distinguishes learning from a teacher’s token probabilities, copying its output tokens and imitating final answers without access to hidden reasoning.

The team works in San Francisco and has investor backing. TypeSafe joined the 2024 AWS Generative AI Accelerator, whose 80 participants were selected from more than 4,700 applications. The program offered each participant up to $1 million in AWS credits, alongside technical mentorship and support bringing products to market.