Anthropic has formed a partnership with the Pratham Education Foundation to introduce a groundbreaking AI-driven formative assessment tool in Indian schools. This collaboration marks Anthropic’s inaugural strategic partnership with one of the nation’s leading education nonprofits.
The focus of this initiative is Pratham’s Anytime Testing Machine (ATM), which utilizes Anthropic’s Claude model. This system is designed to generate questions aligned with the curriculum, grade handwritten responses, and provide individualized feedback. Currently, it is being piloted with 1,500 students from 20 schools, and it is customized for over 5,000 learners participating in Pratham’s Second Chance program, which aims to assist women preparing for India’s grade 10 board examination.
This partnership comes as Anthropic enhances its presence in India, which has become its second-largest market for Claude.ai. Notably, nearly half of Claude’s usage in India focuses on computer and mathematical tasks, illustrating the developer ecosystem’s capability to create production-grade AI systems at scale.
Addressing the Grading Bottleneck
For three decades, Pratham has been dedicated to closing educational gaps across India. Despite the success of large-scale programs validated by randomized controlled trials, such as the Teaching at the Right Level initiative, grading remained a significant operational challenge.
In many classrooms, teachers often oversee 60 or more students, which hampers their ability to offer personalized feedback on practice exams. This issue is especially pressing within the Second Chance program, which assists women who have exited formal schooling and lack access to qualified instructors.
As stated by Nishant Baghel, Director of Technology Innovations at Pratham and a visiting scientist at MIT Media Lab, “These women didn’t have the opportunity to practice the exam adequately because we lacked the staff to grade all those responses.”
The Anytime Testing Machine was developed specifically to overcome this limitation. Students write their answers by hand, take photographs of them, and upload those images. The system then converts the images into text and utilizes Claude to assess responses based on structured rubrics, offering feedback on content, accuracy, and expression.
Pratham chose Claude after reviewing various models. Sravana Chandra, AI Lead at Pratham, explains: “Claude consistently excelled across tasks like question generation, question quality assessment, grading, and feedback. Anthropic’s commitment to safety and responsible AI also influenced our decision to select Claude over other LLMs.”
Iteration and Accuracy Gains
Anthropic and Pratham’s teams convened weekly over several months to fine-tune the grading process. Initially, grading accuracy stood at approximately 30 percent when compared to expert-reviewed benchmarks. However, through iterative prompt engineering and evaluation design, this alignment improved to nearly 80 percent with subject-matter experts.
Chandra elaborates: “We established an LLM-as-a-judge framework, measuring model evaluations against internally produced golden datasets that were manually verified by subject matter experts.”
In terms of question generation, the system now achieves 90 percent accuracy according to Bloom’s Taxonomy. The Claude-enabled ATM has facilitated over 1,500 student assessments across 20 schools, with plans to extend this to 100 schools by the end of 2026. The Second Chance program, which currently supports 15,000 women preparing for the grade 10 exam, aims to fully transition to the Claude-powered pipeline by the conclusion of 2026.
Multilingual functionality has been a crucial aspect of this deployment. Chandra remarks: “Given Claude’s linguistic abilities, it effectively handles feedback generation in both Hindi and English, using English terms where necessary (such as scientific terms) while keeping most of the text in Hindi.”
Keeping Teachers at the Center
A fundamental design principle is that teachers review and can modify AI-generated feedback before it reaches students. Chandra states: “Teachers feel empowered because they are the final evaluators who confirm AI feedback prior to it being delivered to the student.”
Baghel adds: “Human agency is a primary focus for us. Rather than asking what AI can accomplish without teachers, we consider how AI can assist both teachers and students precisely where they encounter difficulties.”
Madhav Chavan, Co-founder of Pratham, believes that “AI tools like Claude provide us with a unique opportunity to transform learning for students who lack advanced educational resources. Beyond offering personalized support for textbook comprehension, the ATM innovation will assist children in verifying and consolidating their knowledge beyond the curriculum.”
Chavan envisions a broader ambition: “Rather than questioning students about the curriculum, we should be asking them about what they know. By shifting this perspective, we can revamp the education system from being merely a filtration mechanism to one that provides differentiated pathways tailored to each child’s interests and foundational knowledge. This was unattainable before the advent of AI.”
Anthropic and Pratham are now broadening their partnership to further support Tech in TaRL (Teaching at the Right Level), with a randomized controlled trial set to involve several thousand students. They are also exploring initiatives related to digital public infrastructure, such as knowledge graphs and potential expansions into Kenya, Rwanda, and other areas in the Global South.
This collaboration aligns with Anthropic’s overarching strategy for India. The company recently established an office in Bengaluru and announced partnerships across sectors including enterprise, education, and agriculture. India now ranks as the second-largest market for Claude.ai, with educational and instructional tasks comprising 12 percent of total usage.
ETIH Innovation Awards 2026
The ETIH Innovation Awards 2026 are currently accepting submissions. These awards celebrate organizations in the education technology sector that demonstrate measurable impact across K–12, higher education, and lifelong learning. Entries are welcomed from the UK, the Americas, and internationally, with entries being evaluated based on their outcomes and practical applications.
