Rossum Aurora
AI model · Language model (LLM) · Text → Text by Rossum
Next-generation transactional LLM for zero-hallucination document automation.
Rossum Aurora is a proprietary transactional LLM built for intelligent document processing. It enables businesses to accurately extract data from invoices and orders in 276+ languages without templates or hallucinations.
Model specs
The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.
- Languages
- English, French, German, Polish, Spanish, Italian, 276+ languages
Category: Document Management
- Sizes and variants
- Proprietary Transactional LLM
- Access
- API and web-based user interface
- Languages
- 276+ languages
- Model type
- Proprietary Transactional LLM
- Supported languages
- 276+
- Pre-trained fields
- 50
- Proprietary Transactional LLM for document understanding
- Supports 276+ languages and handwriting extraction
- Zero hallucinations architecture for high-accuracy data capture
- 50 pre-trained fields for common transactional data points
- Learns instantly from human-in-the-loop feedback
Overview
Rossum Aurora is a proprietary transactional large language model (LLM) designed for intelligent document processing (IDP). It serves businesses by automating the extraction and understanding of data from transactional documents like invoices, purchase orders, and bills of lading without the need for templates. The model is trained on millions of business records and supports over 276 languages, including handwriting. It distinguishes itself by providing high-accuracy extraction with a proprietary architecture designed to eliminate hallucinations. Rossum Aurora is delivered as the core AI engine within the Rossum platform, accessible via API and a web-based user interface for human-AI collaboration.
Document Understanding
- Transactional Document Training: The model is trained on millions of business records to understand complex document layouts without using templates.
- Format Versatility: It extracts data from various formats including PDFs, PNGs, TIFFs, scans, and computer-readable files like JSON, XML, and UBL.
- Data Extraction: The engine processes both structured and unstructured document data, including invoices, purchase orders, and bills of lading.
- Multilingual Support: It supports data extraction and understanding across 276 languages, including handwriting.
AI Performance and Safety
- Zero Hallucinations: The model uses a proprietary architecture designed to reach high accuracy without generating false information.
- Continuous Learning: It learns instantly from user feedback and human-in-the-loop corrections to improve accuracy and confidence scores over time.
- Pre-trained Fields Indexing: The engine comes with 50 pre-trained fields for common transactional data points to accelerate setup.
- Enterprise-Grade Safety: It provides enterprise-grade safety at its core for processing sensitive business transactions.
Integration and Development
- API Access: The model is accessible via API for seamless integration into existing business workflows.
- Web Interface Integration: It powers the Rossum web-based user interface, which is designed for ergonomic human-AI collaboration.
- Custom Extensions Support: Developers can craft custom extensions in Python or NodeJS to tailor the document automation experience.
- Developer Debugging Tools: The platform provides user-accessible logs from extensions to facilitate effortless debugging.
Use cases
- Accounts payable teams automate invoice data extraction across hundreds of languages to reduce manual entry.
- Logistics companies process bills of lading and packing lists with varying layouts without creating templates.
- Customs clearance agents use the model to extract data from unstructured documents to speed up commercial transactions.
- Developers integrate the transactional LLM via API to build custom document automation workflows in Python or NodeJS.
Specifications
- Sizes and variants
- Proprietary Transactional LLM
- Access
- API and web-based user interface
- Fine-tuning
- Learns instantly from user actions and human-in-the-loop corrections
- Languages
- 276+ languages
- Model type
- Proprietary Transactional LLM
- Supported languages
- 276+
- Pre-trained fields
- 50
- Input formats
- PDF, PNG, TIFF, XML, UBL, JSON, scans
- Handwriting support
- Yes
- Learning mode
- Instant learning from user feedback
How it works
- Ingest Documents: Documents are received through various channels such as email, scanners, PEPPOL, portals, or API.
- Classify and Route: The Rossum Aurora engine identifies the document type, removes spam, and splits bundled PDFs into relevant queues.
- Extract Data with AI: The proprietary LLM extracts data from structured or unstructured layouts without requiring templates.
- Human-in-the-Loop Verification: When confidence is low, the ergonomic UI brings in humans to verify data, and the model learns from these actions.
FAQ
What is Rossum Aurora?
Rossum Aurora is a proprietary transactional large language model (LLM) specifically designed for intelligent document processing, trained on millions of business records.
How many languages does Rossum Aurora support?
The model supports over 276 languages and is capable of reading handwriting on transactional documents.
Does Rossum Aurora suffer from hallucinations?
Rossum Aurora is designed with a proprietary architecture that ensures high-accuracy extraction without the risk of hallucinations common in general-purpose LLMs.
Can the model learn from user corrections?
Yes, the model learns instantly from user feedback and human verification to improve its accuracy and confidence scores over time.
Added by AntSpace from public sources (rossum.ai).