EU AI Act · Classification
The European AI Act classifies AI systems used in education as HIGH RISK (Annex III, Section 3a). Semantikmatch is built to meet those requirements.
Annex III · High-risk
Built for AI Act compliance
AI you can trust. Compliance you can prove.
Semantikmatch is built from the ground up to meet the strictest regulatory requirements — EU AI Act, GDPR, and institutional governance. Every decision is explainable. Every piece of data is protected.
EU AI Act · Date of enforcement
EU AI Act
Annex III
High-risk · Education
01/EU AI Act
Compliant from day one.
- Date of enforcement
- Maximum annual fine
- monthsAhead of the market on regulatory compliance.
- AI Act compliance pillars covered.
Compliance pillars
- 01
Risk Management System — integrated, documented, audited.
- 02
Data Governance & Bias Testing — continuous statistical monitoring across demographics, nationalities, and document types
- 03
Technical Documentation — available to supervisory authorities.
- 04
Record-Keeping & Audit Trails — complete log of every AI decision: timestamp, model version, inputs, outcome.
- 05
Human Oversight — every AI decision requires human validation. The human always decides last.
- 06
Accuracy & Robustness Testing
02/GDPR
Your data, safeguarded by design.
At Semantikmatch, data privacy isn't an afterthought — it's a foundational principle.
FranceGermany
ISO 27001-certified centres
Applicant
TLS 1.2+ · in transit
Semantikmatch
AES-256 · at rest
Storage
- 01
Full audit trails for all data interactions.
- 02
End-to-end encryption — TLS 1.2+ in transit, AES-256 at rest.
- 03
Support for Article 22 compliance, including human review pathways and explicit consent mechanisms.
- 04
European data hosting in ISO 27001-certified centres (France & Germany).
- 05
Granular access controls with multi-factor authentication — IAM restricted server access.
- 06
Daily backups with GDPR-compliant retention.
- 07
Quarterly third-party security audits and penetration tests.
03/Trust & Governance
Governance makes AI actually useful.
Well-governed AI isn't just a legal requirement — it's what makes it actually useful. Semantikmatch is built on transparency, accountability, and human control.
Architecture · 3 layers
LAYER 3
Workflow Orchestration
Multi-agent coordination, configurable rules, SIS/CRM/LMS integration. Institutions define their own workflows and escalation paths.
LAYER 2
Specialised AI Agents
Fraud detection (>98% precision, £5-10/candidate vs £50-200 competitors), English level assessment (IELTS correlation >0.85, 50+ accents, 12 languages), Deepfake detection. Each agent is auditable and explainable.
LAYER 1
Trust Infrastructure
DSHAP (explainability), Audit Trails, Human-in-the-Loop, Bias Testing, AI Act Engine. The foundation of the entire system.
- 09:14:02SCOREapp A-2041 · model v3.2 · in: cv.pdf, transcript.pdf → 82/100
- 09:14:02EXPLAINSHAP · grades +18 · experience +11 · language +6
- 09:14:03BIASparity check · nationality, document type → within range
- 09:21:47REVIEWadmissions officer → decision: interview (human)
- 09:21:48AUDITsigned · role: reviewer · institution workflow v4
- 09:30:00MONITORmodel accuracy · drift check → stable
Governance principles
Fairnessdeterministic scoring designed to reduce bias, with continuous monitoring.
ExplainabilitySHAP technology on every ML model, no black boxes.
Human ControlAI assists, humans decide.
Accountabilityclear roles and responsibilities on both the Semantikmatch and institution side.
Continuous improvementmodel monitoring, network effects: each client improves accuracy for all.
EXCLUSIVE PARTNERSHIPS
