IndiaEducation2026 – ongoingPratham International
Anytime Testing Machine for question generation, grading and student feedback
Pratham's grading accuracy rose from 30% to 80% in an AI assessment system for Indian schools
Problem
Pratham runs education programs across nearly every Indian state. It wanted a way to give students frequent, personalized assessments and feedback without teachers writing and marking every test by hand.
Approach
Pratham built the Anytime Testing Machine on the Claude Platform. Claude writes curriculum-aligned questions, checks their quality, grades digitized handwritten answers against rubrics and drafts feedback in a mix of Hindi and English. Teachers review and can edit the feedback before students see it. Pratham worked with Anthropic on prompt engineering to improve grading.
Results
Takeaway
Grading handwritten answers took repeated prompt refinement before it became reliable. Keeping teachers as reviewers of the feedback kept a person between the model and the student.
Categories
- Industry
- Education
- Business function
- Production and quality
- Innovation type
- Process innovationProduct and service innovation
- AI technology
- Generative AI
- Adoption
- Off-the-shelf tool or platform
- Evidence
- Vendor case study
Sources
Source summaries
Anthropic's customer story describes how the education nonprofit Pratham built an assessment system on the Claude Platform. It reports grading accuracy rising from 30% to 80% and 90% accuracy in generating questions aligned to Bloom's Taxonomy. The page carries no date; the Internet Archive first saved it on February 18, 2026.
Full details are in the original sources.
