Marking, feedback and moderation ยท Open source
Assessment needs more than an answer.
Feedbacker helps educators in higher education use AI to help them mark and moderate coursework against a rubric and brief. The AI suggests; you decide every mark, and approve every word a student receives. It is free and open source.
Built, and already used for real marking and moderation by the university educator who develops it.
What it does
Marking, and moderation. Your judgement in both.
Marking a cohort, with feedback
- Import every student's submission at once, from the bulk download of Turnitin or Canvas.
- Mark each submission against the rubric. You choose whether to see the AI's suggested level for each criterion before you mark, or only after you have marked it (blind).
- The AI drafts feedback from your marks and comments. Feedbacker then checks the draft against them, and flags, for example, praise the mark doesn't support, or a missing next step.
- Compare the cohort's feedback side by side, approve exactly what each student receives, and paste or export it.
Moderation: checking a colleague's marking
- Import the sample of marked work you were asked to check, with the original marker's marks and comments.
- Mark each piece yourself against the rubric, with or without seeing the original marks first, and with the AI's own independent reading alongside.
- See where you, the original marker and the AI differ, criterion by criterion, and record your verdict on each piece.
- Approve and export the moderation record: a summary ready to copy into your institution's moderation form.
What it never does
AI assists.
Educators decide.
01
It never gives a mark
The AI suggests a level for each criterion, with its reasons and quotes from the work. Any provisional mark from those suggestions is calculated, kept apart from yours, and never becomes yours.
02
Nothing released without you
You approve exactly what each student receives, and any later change clears that approval. Feedbacker never sends anything to students or to your institution's systems.
03
Every AI contribution traceable
Each suggestion and draft records which AI made it, the instructions it was given and what it was sent, and feedback adapted from an AI draft is labelled as such.
04
Accessible by design
Every screen is measured against checks from WCAG 2.2 AA, the accessibility standard UK universities are expected to meet. Every control can be reached from the keyboard, and it has been used with a screen reader.
Your data
It stays with you, unless you approve it.
Feedbacker is not a website you upload work to: there are no accounts and no Feedbacker server. It runs on your own computer, and your work lives in folders you keep.
Students' names, IDs and contact details in the text are replaced on your computer before anything can be sent. Images in the work, such as charts and photos, aren't anonymised: you look at each one, and choose whether to send it. You approve each anonymised submission, with its images.
Only then, and only when you confirm it, is anything sent to the AI (Anthropic's Claude, with your own API key):
- for its suggested levels: Feedbacker's instructions, the rubric, the anonymised submission and brief, and the images you approved;
- to draft feedback: much the same, without the images, and with your own anonymised levels, marks and comments for that one student, and the guidance on feedback you approved;
- to suggest an edit: the instructions and the rubric, that piece of your feedback with its flags, and your marking of that criterion; not the submission.
Never the names, IDs or original files, nor another student's work. It all goes through a small program on your computer that keeps a log of what it sends, without the text itself. Anonymisation follows rules, and can miss things: your review is the safeguard.
Where it stands
Early, honest, and evaluated.
Marking and moderation are both built. The educator who develops Feedbacker has used each for real, and written up how it went: read the write-ups of marking and feedback and of moderation.
Not yet: an installer; marking in a team, and agreeing standards across markers; AI from providers other than Anthropic; platforms other than Turnitin and Canvas.
Known limits: it works in Chrome or Edge only; it reads typed Word and PDF files, not scans or handwriting; one educator per piece of marking or moderation; marks are interpreted on the UK university marking scale.
Get started
Try it first with the sample coursework.
Feedbacker comes with made-up coursework to practise on. For now, setting it up takes a few technical steps: you need Chrome or Edge, Node.js, a key for Anthropic's AI (paid for by use), and to type a few commands. The setup guide walks you through it; the user guide then takes you through marking a cohort, or a moderation, step by step.