Neuro-symbolic document verification
Read the documents.
Reveal the inconsistencies.
VendueTech Neuro-Symbolic engine combines AI document understanding with explicit domain rules to flag conflicting claims, missing information, and unmet requirements for expert review.
Usable floor area of Unit A: 84 m²
Unit A, residential apartment, 76 m²
Review required: floor area differs by 8 m²
Check whether both documents describe the same unit and use the same measurement basis before relying on either figure.
Recognised and supported by
Use cases
Different industries. The same need to check the evidence.
Review court auction files before you bid.
- Who uses it
- Bidders and investors, law firms, banks and NPL teams.
- Documents
- Court auction notices, land register extracts, court-ordered valuations, and sale decisions.
- Intended benefit
- Focus legal review on conflicting evidence before a bidding decision.
Unit A is offered free of registered encumbrances.
An active mortgage is recorded against Unit A.
Example check: Encumbrance status should be consistent for the same property across the auction file.
New applications require domain configuration, representative documents, and validation.
Product demos
See how knowledge becomes a reviewable finding.
Chapter 1 · Connect domain knowledge
See how domain concepts and rules are organised through explicit relationships.
How it works
Five stages. Then a person decides.
Follow one illustrative scenario: two floor areas for Unit A
-
1Extraction Read relevant information from supported documents. Valuation p. 3: 84 m² Notice p. 2: 76 m²
-
2Orchestration Select the configured processing context. Real estate · Croatia configuration
-
3Knowledge Connect the information to configured domain concepts. Both values → Unit A › floor area
-
4Reasoning Apply explicit checks. One unit, one floor area → conflict
-
5Certificate Present the result and review findings. Review required: floor area
Business value
Estimate the review capacity you could release.
Illustrative estimate
- Document and finding review: 116.7 h
- Exception review: 16.7 h
Adjust assumptions
Defaults are hypothetical modelling inputs, not measured NeSy results or industry averages. Values stay in your browser and are not sent or stored.
Explore scenarios
The benefit depends on how much human review remains necessary and how many exceptions need investigation.
About VendueTech
Built where the documents are hardest.
VendueTech is a Zagreb-based technology company building Bidsale™, a platform for public real estate auctions across Europe. NeSy grew from the need to check, not only read, auction documents spread across court portals, registers and languages. Research is carried out with the University of Zagreb Faculty of Electrical Engineering and Computing (FER).
venduetech.combidsale.aiZICER, Zagreb, Croatia
Governance
Reviewable findings. Explicit checks. Human oversight.
NeSy can support parts of an AI governance workflow. Compliance depends on the complete system and its use.
European Commission: regulatory framework for AIBuilt by a team with experience at
Our team brings experience across global technology, enterprise services, consumer brands, entrepreneurship, and academic research.
Intellectual property
NeSy architecture: patent pending
Our neuro-symbolic architecture is the subject of European patent application EP26191651.4, pending before the European Patent Office (EPO).
Patent application details
- Application number
- EP26191651.4
- Application title
- NEURO-SYMBOLIC SYSTEM AND METHOD FOR MULTI-DOMAIN STRUCTURED DOCUMENT PROCESSING WITH MODULAR SYMBOLIC REASONING AND BIDIRECTIONAL NEURAL FEEDBACK
- Status
- Pending before the EPO. Not granted.
Funding roadmap
Funded R&D today. The next stage under evaluation.
- Grants received Eurostars · DIGIT (World Bank and Ministry of Science and Education)
- Applications submitted EIC Accelerator · EIC Pathfinder
Awards and recognition received
- EIC Pre-Accelerator
Seal of Excellence 2026 
R&D grant
World Bank and Ministry of Science and Education (DIGIT)
VIDI Innovation award
Silver
Startup Connect
3rd place
Money Motion
2nd place
Slush'd
Finalist
Funding and support received
The project is co-financed by the Croatian Agency for SMEs, Innovation and Investments through the Eurostars programme, with funds from the European Union's Horizon Europe programme and the Swiss-Croatian Cooperation Programme. Eurostars is part of the European Partnership for Innovative SMEs. Awards and funding recognise VendueTech's work and R&D; they do not certify product accuracy or endorse the claims on this page.
Current implementation details
- Existing jurisdiction configurations: Croatia, Germany and Portugal. These are configurations, not comprehensive legal coverage.
- Inputs: PDF, scanned images and text, with OCR and structured extraction.
- Rule-based checks, re-checking when problems appear, and a results interface.
- Admin tools for domain concepts, rules and extraction definitions; draft, preview and publish workflow for extraction definitions with version history.
- Web interfaces, API endpoints and a containerised application.
- A demonstration library of six public-source documents, two per jurisdiction.
- Demonstration documents are historical examples, not current title checks or a legally reviewed benchmark.
- Accuracy, rule applicability, coverage and production throughput need further validation.
- Adding a jurisdiction to configuration does not by itself provide document recognition, reviewed rules or operational support.
- The generated report (CCC) describes the information evaluated and checks applied. It is not a legal certificate, regulatory approval or guarantee of factual correctness.
Measure the opportunity on your own documents.
A scoped evaluation can establish review time, extraction quality, missed issues, false alerts, and operating costs for your workflow.
- 01Define one workflowUsers, requirements, expert-agreed expected findings
- 02Evaluate representative documentsIncluding difficult and incomplete examples
- 03Plan integrationResults define the next stage