We’re excited to introduce Few-Shot Learning! 🚀 This new LLM capability allows your AI models to continuously improve without requiring explicit retraining.
For every document you approve in the validation UI, your model learns and adapt to new document formats and variations on the fly.

Try it out and experience a seamless way to keep your models performing at their best!

We’ve rolled out Anti-Hallucination Handling to ensure your AI models generate more accurate, grounded, and reliable predictions. This enhancement reduces the risk of AI-generated hallucinations by reinforcing factual consistency and filtering out uncertain outputs.

What this means for you:

✅ Increased accuracy – Models stay true to real data, minimizing false predictions
✅ Greater trust – Reliable outputs for production-ready AI solutions

This update helps keep your AI models aligned with reality—so you can focus on building with confidence. 🚀

You can now enable Anthropic's Claude LLM alongside Cradl AI’s DocVision 4.0 model. This integration delivers improved extraction results without requiring upfront training and enhances Cradl AI’s context-sensitive extraction, especially for paragraphs of text where only an excerpt is relevant to you.

To enable LLM Assist, go to your Settings > Workspace. You may also configure unique prompts for each data field you want to extract:

Key Takeaways:

• Our new base AI model has been improved, particularly for tabular data and receipts.
• Dates need to be annotated differently if your model is created after April 30, 2024.

DocVision 4.0

The newest version of our base model, DocVision 4.0, has seen major improvements in reading multiline tabular data, making it much more robust at handling complex document layouts. Version 4 also features enhancements to our receipt model and improved understanding of logo typography.

New Features

«Runs» is a new in-app tool that lets users easily manage all of Cradl AI’s core functionalities in one place. In Runs, you can view your run execution history, making it easier to stay organized in Cradl AI. You can also intuitively upload and parse documents without having to pre-configure import and export destinations. Simply drag and drop the documents you want to parse.

You can now get an overview on how far your documents have progressed in your flow. The new view makes it easier to track and debug flows.

Other changes:

• The PDF viewer scrolls to the page where the prediction is found when focusing on a field.
• Squashed bugs related to confidence for line items predictions not showing in the validator view.