AI Impact Summit

Global AI Impact Commons

LLM-based automated grading and personalized feedback for foundational numeracy in offline classrooms

India flag

India

Education

Moderate

Implementing Organisation

Saarthi Education

India, Delhi, South-east Delhi district

Civil Society

Implementing Point of Contact

Garima Yadav

Programme Lead

Contributor of the Impact Story

Carnegie India

Year of implementation

2025

Problem statement

Foundational numeracy learning in India is constrained by a critical feedback gap. In classrooms with high pupil–teacher ratios, teachers are unable to review each student’s arithmetic work daily or provide timely, individualized correction. As a result, students repeatedly practice incorrect logic, allowing learning gaps to harden into long-term habits. Most existing assessment solutions rely on standardization through multiple-choice questions or digital interfaces. These approaches fail to capture step-by-step mathematical reasoning and are impractical in contexts where students lack access to personal devices. This use case addresses the challenge of delivering high-quality, individualized feedback at scale without disrupting paper-based learning. By automating the grading of handwritten arithmetic work and identifying specific logic errors, the system enables teachers to intervene early and accurately, restoring the feedback loop that is essential for mastering foundational numeracy.

Impact story details

Saarthi Education is a non-profit organisation that works to improve basic maths skills among primary school children in India. For the past seven years, Saarthi has worked directly with schools, reaching over 50,000 students, mainly in affordable private schools. The organisation focuses on helping children build strong early arithmetic skills such as number sense, addition, and subtraction. These are the building blocks for all future learning. Saarthi’s programs are designed around everyday classroom realities, large class sizes, limited teacher time, and little or no access to digital devices for students. Through years of on-ground work, Saarthi observed that many children struggle not because they don’t practice, but because they do not receive timely feedback on their work. Teachers often do not have the time to check every notebook regularly, which means mistakes go unnoticed and learning gaps grow. To address this, Saarthi began developing technology tools that help teachers with time-consuming academic tasks like checking student work, while allowing children to continue learning with pen and paper. Saarthi aims to build solutions that are practical, easy to use, and can be adopted widely by schools, NGOs, and public education systems.

AI Technology Used

Machine Learning

Vision-based language models

Key Outcomes

Efficiency

Productivity, Accuracy

Quality Improvement, Knowledge

Skills Impact, Access

Reach, Inclusion

Equity

Narrative Outcome

Saarthi's LLM-based system automates grading of handwritten worksheets, which has helped reduce teacher grading time by over 50 percent while achieving over 90 percent accuracy at the worksheet level. This is particularly useful in high pupil-teacher ratio classrooms, where teachers cannot review each student's arithmetic work daily. Students receive timely, individualised feedback rather than waiting days or weeks.

Impact Metrics

Reduction in time taken by teachers to grade worksheets after the use of LLM-based automated grading

Baseline Value

Time spent during manual grading of all student worksheets Percentage

Post-Implementation

More than 50% reduction in time taken to grade worksheets Percentage

Internal Monitoring·Apr 2025 - Dec 2025

Grading Accuracy of the AI tool (at the worksheet level)

Baseline Value

70 % grading accuracy was achieved by human grading of worksheets

Post-Implementation

Over 90% level grading accuracy was achieved by the AI tool Percentage

Internal Monitoring·Apr 2025 - Dec 2025

Implementation Context

Deployed

Multiple states and regions in India

Primary school students from Grades 1 to 5 from underserved communities

Key Partnerships

Schools, non-profit organizations, and education implementation partners

Replicability & Adaptation

Moderate

1. When used in new regions, the worksheets and grading logic may need to match the local syllabus and language. 2. Teachers will need a short orientation on how to scan student work and use the feedback during regular classroom time.

* The data presented is self-reported by the respective organisations. Readers should consult the original sources for further details.