AI capabilities
AI where you switch it on. Logged wherever it runs.
Eight capabilities scoped to specific tasks — writing, speaking, proctoring, item analysis. Seven workflow assistants that prepare evidence for a human decision. Each is configurable per institution, supervised by a named educator, and logged end to end. Run all of them, some, or none.
Most platforms market "AI everywhere." Institutions need AI they can defend.
It is easy to add AI features. It is harder to add AI that passes a QA office, a Data Protection Officer, a works council, and an academic board — and still helps the educator on Wednesday afternoon. AI acting on learners without named human review makes that conversation harder, not easier.
NUADU takes the opposite approach. AI is unbundled into small, named, task-bounded modules. Each one does a single thing — drafting an item, applying a rubric to a written response, transcribing speech, surfacing an anomaly for a human to look at. None of them decide. None of them act on a named learner. Each one feeds evidence into a step that a named educator owns.
The result is AI you can describe in one sentence to a procurement reviewer — and still get real value from on the Wednesday afternoon.
Eight capabilities. Each scoped to one task.
Content Scanner
Drafts candidate items from your materials — readings, course notes, training documents. Proposes question type, level and a draft rubric. Educator reviews, edits, approves, or rejects.
the differenceEvery item that reaches a learner has been approved by a named educator. AI never publishes an item.
- ✓Human decides
- Approve, edit or reject every drafted item
- ✕Never does
- Publish an item
Demo only; in institutional use the educator reviews, overrides and signs.
AI-assisted marking · written
Applies your rubric consistently across written responses. Surfaces rubric-anchored notes, flags borderline cases, and shows where it is less confident. The marker reviews, overrides where needed, signs off.
the differenceAI is a documented second rater. The educator decides the grade. Every override is logged with a reason and reviewable by QA.
- ✓Human decides
- The educator decides the grade
- ✕Never does
- Override the marker's sign-off
Demo only; in institutional use the educator reviews, overrides and signs.
AI-assisted marking · spoken
Transcribes spoken responses and applies your speaking rubric — pronunciation, fluency, coherence, grammar, vocabulary. CEFR-referenced where applicable.
the differenceThe marker can replay any response, review the transcript, adjust scores, and sign off. AI never finalises a speaking grade.
- ✓Human decides
- Replay, review the transcript, adjust scores, sign off
- ✕Never does
- Finalise a speaking grade
Demo only; in institutional use the educator reviews, overrides and signs.
Pronunciation feedback
Phoneme-level feedback for language learners during practice. Indicates which sounds drift from a target and offers practice prompts.
the differenceUsed in formative practice, not high-stakes scoring. Teachers see the same feedback the learner does and decide whether to act on it.
- ✓Human decides
- Teachers decide whether to act on the feedback
- ✕Never does
- Used in high-stakes scoring
Demo only; in institutional use the educator reviews, overrides and signs.
Long-form review
At draft stage, surfaces structural signals on long-form writing — argument flow, evidence linkage, coherence between sections. Designed for formative feedback, not summative grading.
the differenceOutput goes to the supervising educator. Not a substitute for thesis examination.
- ✓Human decides
- Output goes to the supervising educator
- ✕Never does
- Substitute for thesis examination
Resource search
Semantic search across approved learning content. Finds the segment of a reading, video or document that addresses a concept.
the differenceSurfaces material; the educator decides what to assign. No profiling of the learner.
- ✓Human decides
- The educator decides what to assign
- ✕Never does
- Profile the learner
Integrity signals
A layered set of integrity signals proportionate to exam stakes. Lockdown browser and identity verification at exam start are the defaults. Recorded review and additional integrity layers are opt-in and configurable per assessment.
the differenceSignals are surfaces — not findings. Anomalies are reviewed by a named invigilator or academic-integrity officer, with the candidate's due process intact.
- ✓Human decides
- Anomalies reviewed by a named invigilator
- ✕Never does
- Treat signals as findings
Item analysis & cohort indicators
Item quality statistics per item and aggregated patterns at cohort, class and programme level. Where teaching is landing, where it isn't, where to consider resourcing changes.
the differenceDesigned to inform programme decisions — not to label individual learners or predict their outcomes.
- ✓Human decides
- Programme decisions stay with educators
- ✕Never does
- Label individual learners or predict their outcomes
Seven workflow assistants. Each supervised by a named educator.
Workflow assistants prepare evidence for a human decision. They do not decide. They do not act on a learner. Every output is reviewable, every override is logged.
Marking assistant
Applies your rubric across submissions, flags borderline cases and high rubric-criterion variance for the marker to look at first. Shows where it is confident and where it is not.
- Triggered by: exam submission
- Output goes to: the marker
- Decision: educator signs off
- ✓Human decides
- Final mark on every flagged case
- ✕Never does
- Release a mark on a flagged case without the marker
Standards mapper
Proposes framework alignment — Bloom's, CEFR, EQF, your custom framework — for each item or assessment.
- Triggered by: item creation / import
- Output goes to: the author or QA reviewer
- Decision: alignment approved by a named human
- ✓Human decides
- Accept or change each mapping
- ✕Never does
- Write a mapping without approval
Feedback assembler
Drafts rubric-anchored feedback that the marker can edit before release. Designed to make rubric-anchored feedback feasible at scale without losing the marker's voice.
- Triggered by: marks signed off
- Output goes to: the marker
- Decision: marker approves before release
- ✓Human decides
- Edit and release
- ✕Never does
- Invent criteria outside the rubric
Study-path drafter
After a result, proposes a practice route addressing the rubric criteria where the learner scored lowest. Learner opts in. Drafted from approved content; not a profile — never as labels.
- Triggered by: result released
- Output goes to: the learner (opt-in)
- Decision: learner chooses whether to follow it
- ✓Human decides
- Which steps to use
- ✕Never does
- Make steps mandatory
Optional study companion
A learner-controlled, opt-in companion for practice. Bounded to the learner's recent work and approved content. The learner controls when it is on, sees the data it uses, and can switch it off. Never used as a label.
- Triggered by: learner opt-in
- Output goes to: the learner
- Decision: learner controls scope and use
- ✓Human decides
- What the learner uses or ignores
- ✕Never does
- Draw on sources outside the set
Cohort indicator
Surfaces aggregated indicators at class, cohort and programme level. Designed for resourcing decisions — not individual prediction.
- Triggered by: scheduled / on demand
- Output goes to: programme leads and tutors
- Decision: human resourcing call
- ✓Human decides
- Whether and how to respond
- ✕Never does
- Create a profile of an individual
Programme reviewer
For academic review committees: surfaces patterns across cohorts and programmes over time. Identifies where intended learning outcomes are being met and where they aren't.
- Triggered by: scheduled review cycle
- Output goes to: academic committees
- Decision: human curriculum and programme decisions
- ✓Human decides
- Curriculum or programme changes
- ✕Never does
- Rank or profile individuals
From submission to signed result.
Each stage flows into the next. AI prepares the evidence. The educator decides, approves, signs off. Every output is logged. No automated decisions about a named learner.
Why this is AI institutions can adopt — not just admire.
Modular by design
Each AI capability is a separate, named module. Institutions choose which to run. Everything on this platform works with every AI capability switched off.
Supervised by name
Every AI output that reaches a learner has a named educator's approval behind it. Every override is logged. AI never decides a grade, a flag, or a credential.
Logged end to end
An audit trail your QA office, DPO, candidate appeal process or external auditor can actually read. Inputs, outputs, overrides, approvals, decisions.
Aggregated by default
Programme-level patterns inform resourcing. Individual signals are framed for human conversation — not predictions or labels.
Documented
A documentation pack covering methodology, validity evidence, bias review, retention, data flows and sub-processors is available for institutional review.
Aligned with your rubric
AI applies your rubric, your framework, your standards. Not a generic LLM dressed up. Every output is anchored to your own assessment design.
Questions, answered.
Walk through the AI with your compliance lead in the room.
A 30-minute review of every AI module, the human-accountable workflow, the audit trail, and the documentation pack. Bring your QA officer, DPO or procurement reviewer.