Adaption

Adaption builds tools that help enterprises, startups and research teams shape training data and create custom AI models. Adaptive Data enhances and localizes datasets for particular languages and contexts. AutoScientist jointly optimizes data and training recipes, cycling through training and evaluation toward a user-defined goal. Developers can operate it through a dashboard or API and SDK, then download the resulting model weights to deploy anywhere. This supports organizations whose language or domain requirements are poorly served by general-purpose models.

Co-founders Sara Hooker, CEO, and Sudip Roy, CTO, previously led Cohere’s research division and inference work, respectively. Their focus combines efficient adaptation with systems that retain useful information over time. Adaption’s agent-memory research separates narrative memories, which preserve reasoning and context, from atomic memories that retain precise facts. It also records updated facts alongside the states they supersede, allowing retrieval to favor current information without discarding its history.

A 2026 AI Singapore pilot provides a concrete example of the platform’s use. The company reported generating approximately 750,000 localized training samples across five Southeast Asian languages or regional variants within one month using Adaptive Data. It also reported that AutoScientist saved 56 engineering hours of manual training-hyperparameter optimization in the pilot.