Kagaz AI — From Student Work to the Next Teaching Action

Inspiration

Every student learns at a different pace, but teachers often have limited time to understand each child's learning level in a classroom with diverse needs. Correcting worksheets can reveal which answers are wrong, but identifying the underlying learning gaps and deciding what to teach next can be much harder.

We wanted to explore how AI could help teachers move beyond marks and turn everyday student work into meaningful insights and actionable teaching decisions.

What it does

Kagaz AI is an AI-powered formative assessment and instructional decision-support system focused on foundational literacy and numeracy.

• Capture: Teachers photograph handwritten student worksheets using a smartphone.
• Digitize: OCR extracts student responses for teacher verification.
• Analyse: The system aims to map responses to learning competencies and highlight possible learning gaps.
• Recommend: It proposes flexible learning groups and targeted activities for teachers to review.
• Reassess: Follow-up assessments help teachers review progress and adjust their teaching approach.

Kagaz AI follows a simple principle: Scan → Understand → Teach → Reassess. Teachers remain in control of instructional decisions.

How we built it

We approached Kagaz AI as a teacher-centered, AI-assisted workflow rather than just another grading tool. The project builds on an OCR-based approach to digitizing handwritten student work, with a broader workflow for competency mapping, learning-gap analysis, and instructional recommendations.

The proposed architecture combines image preprocessing, OCR, teacher verification, competency-based analysis, and a recommendation layer. Teacher review is an important checkpoint because handwriting recognition can make mistakes and an incorrect answer does not always reveal the reason behind a student's difficulty.

Our focus is on building incrementally, reusing existing capabilities where possible, and keeping the workflow practical for classrooms that rely on paper-based assessments.

Challenges we ran into

One key challenge is that handwritten responses can be difficult to interpret consistently. OCR errors can affect every subsequent stage, making teacher verification essential.

Another challenge is distinguishing an incorrect answer from an underlying misconception. A single response may not provide enough evidence to determine why a student struggled. Kagaz AI therefore needs to represent uncertainty and allow teachers to conduct additional checks.

We also considered classroom constraints such as limited teaching time, mixed learning levels, device availability, connectivity, and the need to protect children's data. Recommendations must be useful, understandable, and practical—not simply technically impressive.

Accomplishments that we're proud of

We developed a clear vision for connecting student work with instructional action through a continuous formative-assessment cycle.

We are especially proud of the teacher-in-the-loop approach: the goal is not to replace educators, but to reduce the effort required to interpret student work and support better-informed decisions.

We also shaped the project around a practical classroom workflow, bringing together worksheet capture, OCR verification, competency-level insights, suggested interventions, and reassessment within one coherent concept.

What we learned

We learned that applying AI in education involves more than extracting text or generating recommendations. The system must provide useful evidence, communicate uncertainty, and fit naturally into a teacher's existing workflow.

We also recognized the importance of validating recommendations with educators, protecting student information, and measuring practical outcomes instead of assuming that AI-generated insights automatically improve learning.

Most importantly, technology should support teachers' expertise and help them spend more time responding to students' needs.

What's next for Kagaz AI

Our next steps are to strengthen OCR reliability, develop and validate competency mapping, build a library of teacher-reviewed remediation activities, and improve the system's ability to suggest flexible learning groups.

We aim to evaluate Kagaz AI with educators using appropriate permissions and protected student data. Evaluation will focus on OCR correctness, teacher agreement with competency assessments, recommendation usefulness, teacher workload, and changes observed through reassessment.

Longer term, we hope to explore multilingual support, low-connectivity workflows, and responsible deployment across diverse classrooms.

Our vision: Every student's work should help a teacher understand what the child needs next.

Built With

• ai
• computer-vision
• machine-learning
• natural-language-processing
• ocr
• python
• streamlit
• web