Jenna
How we set up AI pre-assessment on your platform
Jenna first-marks your learners' coursework on your own portal, writes feedback the way your assessor does, and leaves the final decision to your assessor. This page explains how we get there: what we do, what we need from you, and how you know the result can be trusted before you rely on it.
Nothing to install · No change to your portal · Your assessor signs off every Pass
Four principles
A co-pilot, not an auto-marker
Regulators and awarding bodies are clear that AI must never be the sole marker. Jenna does the first pass; your assessor reviews and signs off. Nothing is final without a human.
Your standard, not ours
Marks, minima and criteria come from the awarding body's guidance, verbatim. We never author marking criteria. Where the guidance leaves judgement to the assessor, we derive the bar from your assessor's own marking and have them ratify it.
Measure before trust
Before any live use, the standard is applied blind to learners your assessor has already marked and compared criterion by criterion. We look for the pattern behind disagreements, not just an agreement rate.
Protect your platform and budget
Discovery is read-only. Data capture is rate-capped and click-free. Credentials are stored as encrypted secrets. Every run is recorded step by step. Stalled runs are stopped, not left to consume minutes.
Six phases
A first qualification level typically takes two to three weeks elapsed, most of it waiting on learner submissions and assessor time rather than on us.
1. Discovery, read-only
We explore your assessor interface without changing anything: how chapters, questions, marks, feedback and Pass/Refer work, what is deep-linkable, where answers are images or files. We extract the awarding body's assessment guidance and your course manuals to text.
2. The marking standard
One knowledge entry per question carrying the guidance's own wording: the marks available, the minimum, the criteria. Feedback topics are grounded in your manuals so comments point learners to the right material without giving the answer. You review each chapter's standard in a readable document before it is used.
3. Behaviour design
How Jenna acts is agreed with you and written into every mission: comment where marks are lost, record the mark she can justify, never refer on her own, re-review only what the learner revised, record a Pass only when every minimum and the total are met, hand to a human after three rounds. Feedback is always answer-free.
4. Set-up on your Jenna account
One mission per chapter, each carrying the list of learners to assess. A new cohort is the same mission with new names. Credentials are stored as encrypted secrets and revealed only at login. Marking uses the high-effort model mode; narration is off; each chapter has a step budget sized to its length.
5. Validation against your assessor
The standard is applied blind to learners your assessor has already marked and compared criterion by criterion. Live runs are second-marked independently. We look for the direction of disagreement and bias the standard towards strictness, because a stricter first pass costs a learner one revision while a lenient one costs you a false Pass.
6. Calibration and go-live
Your assessor's own marks and feedback are read from the portal with plain, rate-capped page requests and no clicks. From them we derive a calibration layer for the criteria the guidance leaves vague, kept separate from the guidance text, and your assessor ratifies it. We re-run the questions that were over-marked, on empty mark boxes, and confirm they move. Then learners are first-marked as they submit.
What stays human
- Every Pass. Jenna submits under the assessor account; your assessor reviews and remains accountable. The errors that matter live at the pass boundary, so that is where the human looks.
- The bar for vague criteria. "A more detailed answer" means what your assessor says it means. We derive it from their marking and they ratify it.
- Learners who do not converge. After three comment-and-revise rounds on the same answer, the chapter is flagged for a person.
- Practical and in-person assessments. Only what is marked online is in scope.
Safety and data handling
- Read-only discovery. Nothing on your portal is changed until the first agreed live run.
- Rate-capped capture. Reading your assessor's marks for calibration uses plain page requests at one request per second, with no clicks and no AI in the loop.
- Credentials as secrets. Logins are stored encrypted on your Jenna account and never appear in mission text or logs.
- Audit trail. Every run keeps a video, a step-by-step log with screenshots and the rationale behind each mark.
- Anonymisation. Learners are never named in anything that leaves your account.
- Budget control. Minutes are tracked to the decimal; runs that stall are stopped; nothing polls on a schedule.
What it costs
Set-up is billed by the hour and quoted after discovery. Running costs are Jenna automation minutes on your own account. On our reference engagement, a first-marked chapter consumed roughly 30 to 50 minutes of browser time in high-effort mode.
| Marker | Time per chapter | Cost per chapter |
|---|---|---|
| A qualified assessor | 20 – 40 minutes | ≈ €7 – 13 |
| Jenna, standard model | ≈ 36 minutes of browser time | ≈ €1.4 |
| Jenna, high-effort model | 30 – 50 minutes of browser time, billed at 2.5× | ≈ €3.5 |
Figures measured on Active IQ Level 2 chapters of 14 to 16 questions. Your chapters may be shorter or longer.
Questions centres ask
Do we need to change our learning platform?
No. Jenna works through a real browser on your existing portal with a dedicated assessor account. There is no integration, no plugin and nothing to install. If your portal is deep-linkable, runs are faster; if not, Jenna navigates like a person would.
Does the AI decide who passes?
No. Jenna first-marks and leaves feedback. Where an answer is below its minimum, the chapter is held with comments for the learner, never referred. Your assessor reviews every result and remains accountable for the outcome.
What about answers that are images, posters or PDFs?
Jenna opens the preview your portal provides and marks from what she sees; a poster assignment was marked 13 out of 13 in agreement with the assessor on our reference engagement. Hand-annotated images and anything the portal cannot preview are flagged for a person.
How do you keep learners' data safe?
Jenna runs on your own account with your own credentials, stored encrypted. We read your portal only as agreed, at a capped rate. Learners are anonymised in any report that leaves your account. Run recordings live on your account and can be deleted on request.
What happens when the model is wrong?
Every mark carries its rationale and the learner's quoted words, so a wrong mark is visible and correctable in the portal exactly as today. Systematic errors are what validation and calibration exist to find; on our reference engagement the pattern was leniency on vague criteria, and the calibration moved seven of nine targeted questions to the assessor's mark.
Which qualifications and platforms?
Any written coursework marked online against a published mark scheme. We started with Active IQ Level 2 and 3 on a custom Laravel portal; Moodle and similar learning platforms follow the same six phases.
Start with a 15-minute discovery call
Tell us which qualification you mark and on which platform. We will say honestly whether it fits and what the set-up would take.