Home / smartextract
Private Sector
Stage 1, Stage 2

smartextract

dida Datenschmiede GmbH
‹ Back to Open Call Winners
smartextract by dida (Berlin) turns unstructured business documents such as invoices, order confirmations and delivery notes into validated, structured data for ERP, CRM or LIMS, with no templates and no training. Open-weights LLMs read any layout in 80+ languages, rules and ERP master data check every field, and humans review only flagged items. In production for 3+ years with 55,000+ documents per month: up to 80% faster processing and 95% less manual work. Hosted in the EU or on-premise, it brings a sovereign European alternative to US hyperscaler tools onto AIoD.

Background

Every European company runs on documents nobody wants to type. Wholesalers, manufacturers, logistics providers and public bodies receive order confirmations, invoices and delivery notes as PDFs, scans or e-mails, and every supplier uses its own layout and changes it without notice. Affected employees lose 30–100 workdays per year re-typing this data into ERP systems. Error rates on manual transfer reach 5–15%: wrong prices, quantities and part numbers cause costly downstream mistakes. Template-based OCR breaks on every new layout, supplier or language, so growth means hiring more data-entry staff instead of becoming more efficient.

Proposed solution

smartextract is a production-grade Document Understanding platform by dida, a Berlin-based AI company. It converts unstructured documents (PDF, TIFF, JPEG, PNG, e-mail) into validated, structured data (JSON, CSV, XML) and transfers it directly into ERP, CRM, LIMS or Excel via a REST API. The pipeline splits bundled files, classifies each document, extracts header and line-item data, validates it and exports it. Layout-aware parsing reconstructs the page structure. Open-weights multimodal LLMs (Mistral and Llama, fine-tuned with LoRA) then interpret it semantically, so new layouts work on day one without templates or training. A deterministic validation stage checks every record against business rules (e.g. quantity × price = total) and ERP master data, which neutralises hallucinations. Every field carries a confidence score, and only flagged fields go to a human reviewer, for up to 99.99% export accuracy. smartextract runs at TRL 8/9 and processes 55,000+ documents per month in 80+ languages. It is hosted in the EU or on-premise and is being onboarded as a containerised, API-first service on the AIoD Marketplace.

Objectives

  • Automate template-free extraction of line-item data from unstructured business documents (order confirmations, invoices, delivery notes) in 80+ languages
  • Cut manual verification workload by up to 95% and verification time by over 80%
  • Flag deviations in price, quantity and delivery date against ERP data
  • Ensure reliable output through business-rule validation, confidence scores and human-in-the-loop review
  • Provide a GDPR-compliant European alternative built on open-weights models and EU hosting
  • Enable easy integration via REST API and standard JSON/XML/CSV output

dida Datenschmiede GmbH

Germany
dida is a Berlin-based machine learning company with eight years of experience delivering AI systems for enterprises. A team of 35+ mathematicians and physicists has completed more than ML projects specialising in NLP and information extraction, computer vision, semantic search and quality control. dida serves clients across industry and manufacturing, energy and infrastructure, trade and logistics, and the public sector – with a privacy-first approach based on EU hosting or on-premise deployment. Its work has been recognised with the AI for Earth Award and Global Top 100 AI Projects listing.
Stay informed on AI opportunities
Get updates on new solutions, market trends, and adoption strategies delivered to your inbox.
The DeployAI project has received funding from the Digital Europe Programme (DIGITAL) under grant agreement №. 101146490 - DIGITAL-2022-CLOUD-AI-B-03
Stay informed on AI opportunities
Get updates on new solutions, market trends, and adoption strategies delivered to your inbox.
The DeployAI project has received funding from the Digital Europe Programme (DIGITAL) under grant agreement №. 101146490 - DIGITAL-2022-CLOUD-AI-B-03