Edundy’s New and Improved AI Grader Is Live: Faster, Smarter, and More Reliable Than Ever

Edundy is committed to building grading tools that make assessment more accurate, more consistent, and more manageable for teachers. This week, we launched a major upgrade to our AI grading engine, now available to all schools using Edundy. This update represents one of the most substantial improvements to the platform since our initial release, bringing higher precision, stronger reliability, and clearer feedback for students.

Below is a detailed overview of what has changed, why these updates matter, and how educators can get the most out of the new grading system.

Major Improvements to Accuracy and Reliability

The foundation of this update is a redesigned grading engine with significant enhancements in several core areas.

2X More Consistency

Consistency is one of the most important elements of fair and dependable grading. The upgraded system now assigns the same score to the same submission every time. This reduces variation between grading sessions and helps teachers feel more confident that the results align with the rubric they designed. Consistency also supports better longitudinal tracking, since teachers can trust that improvements in student work reflect genuine growth, not fluctuations in the scoring model.

3X Better Handwriting Recognition

Handwriting remains one of the largest challenges in automated grading, especially when students rush or when younger learners are still developing fine motor skills. The improved handwriting recognition engine has been trained on a much broader and more varied set of writing styles, allowing it to interpret unclear or messy handwriting more accurately. This leads to fewer misreads, fewer incorrectly identified words or symbols, and more accurate evaluation of student intent.

3X Reduction in Hallucinations

“Hallucinations” occur when an AI infers details that were not actually present in the submission. Reducing these errors is essential for fair assessment. The upgraded model is far more grounded in the text and visuals students provide. Instead of making assumptions, it relies strictly on what is observable in the submission. This improves trust and ensures students are assessed on their actual work, not on accidental interpretations.

Improved Handling of Messy or Large Work

Real classroom submissions rarely arrive in perfect condition. Pages may be out of order, margins may contain stray notes, and work may span multiple images taken at different angles. The new grader is more robust in handling these situations. It can follow disorganized work more effectively, interpret steps that may not be cleanly aligned, and analyze longer multi-page submissions with more stability. This upgrade supports a more realistic range of classroom workflows.

Stronger Graph and Diagram Analysis

Subjects like math and science require students to draw coordinate planes, label axes, sketch diagrams, and construct graphs. These visual components are often critical to demonstrating understanding. The updated grader has a significantly improved ability to interpret graphs, identify key components, and evaluate whether the student built or labeled the visual correctly. This helps provide more accurate scoring in assignments that rely heavily on visuals.

Clearer, More Actionable Feedback for Students

One of the goals of this update is to help students understand not just their score but why they received it. Feedback is now more direct, more aligned with rubric criteria, and easier for students to interpret independently. While additional enhancements to feedback are in development, the current version already produces more structured explanations and more meaningful guidance for improvement.

Why the System Feels More Literal

One major change educators will notice is that the grader interprets rubric language much more literally than before. Every part of the rubric is analyzed, including small details, specific wording, and implicit expectations. This literal approach improves consistency and fairness, but it may lead to slightly lower scores in some classes until teachers adjust wording to match their instructional intent.

Literal interpretation means:

• Every criterion must be met exactly as defined.

• Subtle wording choices can influence scoring outcomes.

• High-performance levels must clearly state what is required and what amount of deviation is acceptable.

  
  

Teachers often design rubrics with assumed flexibility, but AI does not assume. This update ensures the system evaluates rubrics exactly as written, which may highlight areas where expectations need refinement.

Recommendation: Review the Top Levels of Your Rubrics

To get the most out of the new grader, we recommend reviewing the highest performance level of your rubric. If full credit currently demands flawless execution, the grader will apply that requirement strictly. In practice, most teachers allow for minor mistakes that do not affect overall understanding. For this reason, adjusting rubric language can help the system reflect common instructional practices.

Here is an example:

Original rubric language: “Student shows complete understanding with no errors.”

Improved version: “Student demonstrates complete and accurate understanding; minor errors that do not affect conceptual understanding are acceptable.”

Small adjustments like this clarify the criteria and make grading more aligned with instructional intent.

Why This Upgrade Matters for Schools

This release is designed to reduce time spent reviewing inconsistencies and correcting misreads, while increasing trust in automated scoring. By improving the model’s ability to interpret handwriting, diagrams, large submissions, and messy work, Edundy ensures that teachers spend less time checking for AI mistakes and more time teaching.

Stronger feedback also supports student growth and helps learners engage more deeply with their results.

What’s Next

This upgrade is the foundation for several upcoming enhancements focused on feedback quality, rubric analysis, and teacher customization. We are continuing to invest in research and development to extend the grader’s capabilities and deliver even clearer and more personalized insights to students.

Questions or Support?

If teachers or administrators have questions about the new system, want assistance adjusting rubrics, or notice anything unexpected in grading behavior, the Edundy team is ready to help. We welcome feedback and are committed to building tools that strengthen teaching and learning across every subject area.