LaTeX, Word, LibreOffice, Markdown or AsciiDoc: keep your usual tool, just add the alignment marks and the identity boxes.
Copies are straightened and aligned on the template in your browser, or, for large batches, by a dedicated server worker that you can follow live and leave running.
Handwritten names, first names and student IDs are recognised to match each copy with a student, and the grades are exported in the format expected by the university administration.
In order to be able to correct in a horizontal way (all the questions n°1, then all the questions n°2), with the page count of each scan checked against the template.
Direct, positive, negative or hybrid scoring (with bonus and penalties) to fit the question you are asking, and statistics per question.
Ticked boxes are detected automatically. A small grading language gives partial credit or penalties (a&b=100; a=50; *=-25), and your manual corrections are kept when you run the autocorrection again.
Re-apply your comments in 1 click on another student sheet, or grade a whole group of similar answers at once.
Annotate the sheets with your pen, move between copies with a touch pad or a swipe, rotate an answer zone by 90°, and focus with the zen mode.
In order to generate a sample answer sheet for the students and discuss the wrong answers with the class.
Students enter their e-mail address and receive a temporary link to all their copies. You choose, per exam, what they see: nothing, the scanned copy, or the graded copy with your annotations. Annotated PDF exports are also available.
Give qualitative feedback on Jupyter notebook assignments with the same correction tools.
Use the hosted instance, deploy your own with Docker or Kubernetes, or run the standalone version on your laptop. Login with CAS or Shibboleth for universities.

Several Master 1 projects are working on a student exam assessment tool. Architecture on top of Eclipse RCP (Java + SWT).

Didactic support project for several courses of the Computer Science Master of the University of Rennes. The code is not written by the students but we take excerpts of the code to introduce the DevOps challenges, WebAssembly, the notion of Web Worker, ...

A teaching module in electronics, a teaching module in computer science (Master MIAGE), first feedbacks.

Some new features are added: LaTeX, Word and LibreOffice templates for the creation of exam statements, translation update, recruitment of new testers, ...

New tests at Univ Rennes and Univ Toulouse, support for MCQ, bug fixes, ...

Improved ergonomics, performance and architecture, University IS integration, over 85 pull requests since the beginning of January ...

Grade groups of similar answers at once (CorrectExam automatically forms groups for you to review), improved performance, module import/export, standalone version (to host everything on your own laptop), ...

Scalability improvements, PDF manipulation (rotating pages, moving pages), ...


New zen mode for correction, random correction, LaTeX template evolution, improvements based on feedback from users...

Recognition of handwritten texts. Transformer architectures deliver low error rates but are challenging to train due to limited annotated data in handwritten text recognition. AI is still executed in the browser.

nbgrader is a tool that facilitates the creation and grading of assignments using Jupyter notebooks, as well as the distribution and retrieval of these assignments to students. It allows teachers to easily create notebook-based assignments that include programming exercises, explanatory text, images or videos, and free-form written responses. The integration with CorrectExam allows teachers to use CorrectExam for more qualitative feedback to students.
Upgrade of the front-end dependencies and of the ONNX Runtime used for handwriting recognition, improved English and Portuguese translations, cache management fixes.
First version of deskewandcachecreator, an alignment worker written in Rust (PDFium, AKAZE, corner disk detection) fed through RabbitMQ and S3, so that very large scans no longer depend on the power of the teacher's laptop.
Server-side alignment available from the alignment screen, with live progress and the ability to come back to a running job; a student portal where students get all their copies through an e-mailed link, with per-exam visibility chosen by the teacher; a grading language for MCQs (partial credit, penalties) and manual MCQ corrections preserved; touch navigation pad, swipe and 90° rotation of answer zones on tablets; page count of each scan; security and reliability hardening.
Heavy computations stay out of the API server: image processing and AI run in the browser, in Web Workers, and large scans can be delegated to a dedicated worker.
Teamwork is essential. If you make a mistake, you can blame someone else 😀.