tetrel Logo tetrel Logo

docuduct

AI development even without perfect training data

In the real world, training data for AI models is rarely perfect. With the docuduct platform, you can still develop high-performance text analysis models - through a unique combination of training data optimization and model development.
Up to
80%
less development effort
Up to
70%
faster prediction
Exactly
100%
optimized workflow

Because your data is not perfect

The quality of training data is the biggest lever for performant AI models. In reality, however, perfect data is rarely found.

The docuduct development platform turns a few examples into high-performance text analysis models - through an end-to-end workflow that continuously corrects and optimizes the training data.

checkmark
Reproducible, efficient workflow without media breaks
checkmark
Training with a few examples
checkmark
Optimized processing of long documents
checkmark
For individual users and teams
checkmark
Cloud-based or on-premise
checkmark
OCR integrated in >100 languages
checkmark
Connection to common DMS and RPA systems

Suitable training documents in record time

  • OCR with layout recognition in > 100 languages
  • Integrated document conversion supports PDF, image documents, Microsoft Office, JSON and text
  • Generative language for synthesizing training data
  • Full document history with “time-travel” feature

Data annotation with a fraction of the effort

  • Optimized annotation interface for long documents
  • Active learning for outstanding results with up to 80% less annotated data
  • Fast training-annotation-retraining workflow without media breaks
  • Multi-user functionality with annotation guidelines

One-click model training and model management

  • Latest generation transformer models (e.g. BERT) available in various languages
  • Auto-scaling GPU training with one click
  • Model results and model comparison in one convenient interface
  • MLflow integration for detailed training statistics

Automated deployment and optimal inference performance

  • Prediction service with documented REST API at the push of a button
  • Up to 70% faster inference time through automatic model quantization and serialization
  • Multilingual OCR included by default
  • Deployment as Docker container for maximum compatibility

Implementation

Designed for maximum flexibility

The docuduct platform is consequently developed on the basis of standardized interfaces and containers. Whether as a cloud solution or on-premise - it can flexibly be deployed and integrated independently of existing system landscapes.

platform architecture
Cloud or on-premise
Process your data wherever your business demands it: in the cloud or strictly on-premise.
Flexible integration
Use docuduct modules that have already been developed in existing services via API integration or in common document management and workflow systems.
Scalable
Scale with your requirements: from the first proof of concept to a cluster with several hundred real-time services.