The new Buildsimple Foundation model

A consistent further development for stable document processes

December 3, 2025
Marius Gotschol

At Buildsimple, we are developing our AI platform step by step. Our goal is to create a solution that works reliably, can be easily integrated into existing processes, and does not require large amounts of data. Foundation models make an important contribution to this.

What is a foundation model?

A foundation model is a pre-trained base model that has developed a fundamental understanding of language and structure through large amounts of text, documents, and layouts. It recognizes contextual relationships, typical document structures, and patterns that frequently occur in invoices, forms, emails, or contracts.
This pre-learned knowledge makes foundation models more stable and significantly less dependent on large training data sets. This is precisely why they are becoming increasingly popular in document processing.

With the new Buildsimple Foundation model, we are creating an improved technical basis for classification, extraction, and future separation functions within the platform.

Why foundation models are crucial now

Document processes are changing. Today, companies receive data from a wide variety of sources and with highly varying structures. Traditional machine learning approaches often reach their limits, especially when there is little training data available or documents are not structured uniformly.

Foundation models provide a stable basis for this. They are pre-trained on large amounts of text, layouts, and examples and provide a basic understanding that also helps with new, unfamiliar, or incomplete documents. For input management, this means greater robustness, fewer manual rules, and less training effort.

The new Buildsimple Foundation model at a glance

The new model is a proprietary base model developed from extensive document training. It understands language, layout structures, and thematic relationships and can classify new content more quickly.

Key objectives of the new Buildsimple Foundation models:

A better basic understanding of complex document structures

Less dependence on large training data sets

More stable results with varying layouts

Less effort required for setup and customization

A uniform basis for the platform's central AI functions

This means that the model can work reliably even if documents look different, have different layouts, or are of varying quality.

Specific benefits for companies

Companies benefit directly from the model's improved capabilities.

Particularly relevant advantages:

Significantly less training effort
Greater accuracy in classification
Consistent results despite different layouts or scan qualities
Faster adjustments and iterations
Fewer manual rules and reduced maintenance requirements
Better results even with small amounts of data
Improved processing of heterogeneous document stacks

Technical principles and mode of operation

The Foundation model combines various modern methods to better understand and reliably process documents. Each method contributes to meaningfully linking language, layout, and content.

Key components of the model:

Pre-training with many real documents to recognize typical patterns and structures

A shared vector space for text, images, and layout, enabling the model to organize entire pages

A modern AI approach that recognizes relationships within a document

A learning method that reliably identifies differences and similarities between documents

Fine-tuning that enables good results even with just a few examples

The model strengthens both the work in specialist areas and the requirements of IT and architecture.

Applications in the Buildsimple product

The Foundation model forms the technical basis for several components of the Buildsimple platform.

Areas in which the model is or will be used:

Classification: The new engine is already fully based on the Foundation model.
Extraction: The next generation of extraction models will also use the Foundation base.
Document separation: A new separation engine is under development and will use the model for context-based boundary detection.
Migration of existing models: Existing customer solutions can be converted to the new model upon request.
Long-term product strategy: The model serves as the central basis for a wide variety of AI functions in Buildsimple.

Conclusion

The new Buildsimple Foundation model is a logical further development of our AI platform. It makes document processes more stable, accurate, and easier to use, while also forming the basis for future improvements. For companies, this means less effort, higher quality, and more reliable results in their daily work with documents.