Responsible AI at Integrevise
Integrevise builds AI-assisted oral assessment tools for education. Our goal is to help institutions deliver fairer, more consistent, and more evidence-rich viva experiences while keeping academic responsibility with educators and institutions.
This statement explains how we approach trustworthy and responsible AI across the Integrevise platform.
Our Commitment
We believe AI in education should support human judgement, not replace it. Integrevise uses AI to assist with questioning, transcription, feedback, rubric-aligned scoring suggestions, and evidence-rich assessment workflows. Formal academic responsibility remains with the institution, its educators, and its approved assessment processes.
Our responsible AI commitments are:
- Human-led assessment
- Clear and purposeful AI use
- Privacy and data protection by design
- Fairness, accessibility, and learner agency
- Safety, reliability, and security
- Accountability, auditability, and continuous improvement
How Integrevise Uses AI
Integrevise may use AI and related automation for:
- Viva planning and question sequencing
- Live AI examiner interaction in practice, reflective, and assessed viva modes
- Speech-to-text transcription of learner responses
- Text-to-speech examiner voice output
- Retrieval of relevant context from submitted work, assignment briefs, and approved learning materials
- AI-assisted scoring suggestions, rubric-aligned feedback, and evidence summaries
- Formative practice feedback to help learners prepare
Integrevise does not build, train, or fine-tune its own foundation models. Instead, the platform combines third-party AI services with Integrevise-owned orchestration, prompt governance, assessment workflows, rubrics, evidence records, and institutional controls.
Integrevise is not designed to make automatic misconduct findings or act as a standalone plagiarism or AI-detection system. Its purpose is to help institutions verify understanding, support human review, and produce clearer evidence around learner performance.
Human Oversight
AI outputs in Integrevise are designed to be reviewed, governed, and used within an educational workflow.
For formal assessed vivas:
- Institutions and educators define the assessment context, viva configuration, rubric, release policy, and learner guidance.
- AI-assisted scoring and feedback are structured around defined criteria, such as understanding, critical thinking, communication, and subject-specific expectations.
- Risk signals such as low confidence, borderline scores, incomplete responses, or unusual grading outcomes can require manual review.
- Educators and institutions retain responsibility for reviewing, publishing, moderating, appealing, or otherwise acting on assessment outcomes.
Practice feedback is formative. It is intended to help learners prepare and should not be treated as a final academic decision.
Transparency
Learners and educators should understand when and how AI is being used. Integrevise supports transparency through:
- Product and session experiences that identify Integrevise as an AI-assisted viva platform
- Educator-controlled viva settings and assessment criteria
- Session transcripts and records of issued questions and captured responses
- Rubric-aligned feedback and assessment reports
- Manual-review indicators where AI output requires closer human attention
- Audit-friendly evidence that institutions can use for review, moderation, appeal handling, and quality assurance
Current AI systems are not fully interpretable. Integrevise does not claim that large language model outputs are technically explainable in the same way as a simple rules-based calculation. Our approach focuses on process transparency: clear assessment criteria, retained evidence, reviewable outputs, and human accountability.
Privacy and Data Protection
Integrevise processes learner, staff, and institutional data only to provide, secure, support, and improve the contracted service, subject to the applicable customer agreement and data processing terms.
Our privacy and data protection approach includes:
- Tenant-scoped processing and institution-level data separation
- Role-based access controls for learners, educators, administrators, and authorised platform staff
- Data minimisation in AI workflows, so prompts and retrieved evidence are limited to what is needed for the task
- Encryption in transit and protection of sensitive stored values
- Retention, erasure, legal hold, and subject-access export workflows where available under the institution’s configuration
- Sub-processor governance for third-party AI, infrastructure, and media services
Customer and institutional data is not used to train Integrevise proprietary foundation models, because Integrevise does not train such models. Where third-party AI services process data for inference, this is governed by the relevant customer agreement, sub-processor terms, and provider controls.
Fairness, Accessibility, and Inclusion
AI can perform unevenly across people, accents, disciplines, languages, and contexts. Integrevise treats fairness as an ongoing operational duty rather than a one-time claim.
We support fairness and inclusion by:
- Aligning AI-assisted assessment to educator-defined rubrics
- Preserving transcripts, responses, feedback, and assessment records for review
- Supporting manual review where confidence or outcome patterns indicate risk
- Providing language and voice preferences where available
- Supporting institution-led moderation, appeal, and quality assurance processes
- Designing access controls so learner evidence is available only to authorised roles
No AI assessment output should be treated as infallible. Institutions should continue to apply their academic regulations, accessibility duties, reasonable adjustment processes, learner support policies, and moderation procedures.
Safety, Reliability, and Security
AI systems can make mistakes. Integrevise designs controls around that reality.
Our safeguards include:
- Structured output formats for AI-assisted scoring and feedback
- Prompt and workflow governance
- Confidence and manual-review signals
- Fallback behaviour when AI calls fail or produce incomplete results
- Audit records for key assessment and compliance workflows
- Secure software development, dependency review, secrets management, and controlled production deployment practices
Security and responsible AI are connected: trustworthy AI depends on protecting the systems, data, and workflows around the model.
What Integrevise Does Not Do
Integrevise does not:
- Train or fine-tune proprietary foundation models on learner or customer data
- Sell learner, staff, or institutional data
- Treat AI outputs as inherently final or unquestionable academic decisions
- Use AI to infer protected characteristics outside the configured educational workflow
- Claim that AI outputs are always correct, unbiased, or fully explainable
- Replace institutional responsibility for assessment design, moderation, appeals, accessibility, reasonable adjustments, and academic standards
- Act as a standalone plagiarism, misconduct, or AI-detection decision system
Continuous Improvement
Responsible AI changes as technology, regulation, and educational practice change. Integrevise reviews its AI practices through product development, testing, monitoring, institutional feedback, and changes in relevant legal and sector expectations.
We will update this statement when there are material changes to our AI use, governance approach, or responsible AI commitments.
Questions about this statement can be sent to info@integrevise.com.