At Assessain, we believe teachers’ time is too valuable to be spent on repetitive grading. That’s why our platform combines cutting-edge AI models with practical classroom workflows — making grading faster, more consistent, and more insightful.

Here’s a peek under the hood at the AI powering Assessain:

🖊️ Checkbox Detection with a Custom Keras Model

Many paper-based tests include multiple-choice questions with checkboxes. We’ve built and trained our own Keras-based computer vision model specifically to detect these checkmarks, even on scanned or slightly skewed pages. This ensures accurate recognition and saves teachers from manually reviewing every response.

🌍 Multilingual Transcription via Google Document AI

For written answers, we rely on Google Document AI to transcribe student submissions in most languages. This gives us a highly accurate digital version of each response, ready for instant AI evaluation.

💬 Evaluation & Feedback with OpenAI GPT-4o

Once responses are transcribed, OpenAI’s GPT-4o steps in to:

• Evaluate open-ended answers based on teacher-defined criteria
• Provide clear, formative feedback students can act on
• Transform teacher-uploaded materials into structured, scannable worksheets

GPT-4o has been a game-changer in combining speed, reliability, and consistent grading quality.

🧪 Experimenting with GPT-5

We’re also testing the brand-new GPT-5 model. Early results show it’s:

• Slower than GPT-4o
• More prone to over-reasoning (sometimes adding criteria the teacher didn’t request)
• A bit more random in scoring

That said, we’re running structured comparisons of GPT-4o and GPT-5 to measure evaluation quality and feedback clarity. Our goal is to adopt the best possible AI blend for teachers.

The takeaway:
We’re not just using AI for the sake of it — every tool in Assessain is chosen, trained, and tested to make grading faster, more consistent, and more helpful for students. As AI evolves, so will we.