Featured Story

Product

# Expand: Automatic Data Coverage for Customized Model Training

[Adaption Research Team](https://adaptionlabs.ai/research)

·

Nov 12, 2026

·

2min read

![Expand: Automatic Data Coverage for Customized Model Training](https://adaptionlabs.ai/_next/image?url=https%3A%2F%2Fcdn.sanity.io%2Fimages%2F5fn3so8t%2Fproduction%2Ff9bc2cb381d709e95a0778e013062cca5b47a0e9-1780x866.jpg&w=1920&q=75)

Today we introduce Expand, a new feature in [Adaptive Data](https://adaptionlabs.ai/adaptive-data), which allows you to go from a handful of examples of the capability you care about, to a large scale training dataset. Start with 10 examples, and in a few clicks expand to 100,000 data points similar to your tasks and optimized for AI training.

### Expand Adds Examples Your Data Is Missing

![Blog section illustration](https://adaptionlabs.ai/_next/image?url=https%3A%2F%2Fcdn.sanity.io%2Fimages%2F5fn3so8t%2Fproduction%2F47f66dd13e9590e78d43a585fa32a2f9f9d99193-1780x1002.jpg&w=1920&q=75)

Models trained on too few examples often fail on phrasing and situations they haven’t seen. This can lead to hallucinations or inaccurate answers. Expand generates the additional examples a model needs to respond to variation, and supercharge your enterprise context.

If you’re training a model to handle medical insurance claims questions, and you have a few thousand reviewed claims, Expand can grow your dataset in three ways.

1. Add more data like yours. Expand generates synthetic examples in the style of your own data. It samples your examples and closely matches their topics, language, and format. If your data has claim summaries, Expand writes more, across different damage types, amounts, and phrasing.
2. Add related tasks from your domain. Expand adds other tasks within the same domain. For example, other tasks in the medical field. This builds model robustness, so fewer tasks are unfamiliar to your model.
3. Add other domains for optimal training balance. Training quality improves when the data is varied. Turning this on activates a data agent that picks the mix of topics and sources that will maximize performance on your task.

Expand sits alongside Translate and Localize in Adaptive Data. Translate carries your dataset into up to 242 languages, and Localize matches it to a specific language and country, so your model serves every market you work in.

### From Dataset to Frontier Model

Model quality is limited by how closely your data matches the objective you want. Expand brings your data closer to that objective, and [AutoScientist](https://adaptionlabs.ai/auto-scientist) carries it the rest of the way. Bring your expanded dataset into AutoScientist, and it optimizes your data and training recipe together on the open weights model you choose. That cuts the time to a high-performing model from weeks to days.

Expand joins [Invent a Dataset](https://adaptionlabs.ai/invent-a-dataset) as a starting point on the path to a custom model. Invent a Dataset starts from a description when you have no examples at all. Expand starts from the examples you already have, however few. Either way, you end up in the same place: a model trained on your knowledge, fitted to your goal, and owned by your team.

The most valuable intelligence isn’t the biggest model. It’s the one that knows your business and keeps improving as your business evolves. The knowledge that sets your company apart already exists. Expand gives it the scale to train your own custom model.

[Try Expand](https://adaptionlabs.ai/app/auth)[Talk to an AI Expert](https://adaptionlabs.ai/book-a-call)

Author

[Adaption Research Team](https://adaptionlabs.ai/research)

Date

Nov 12, 2026