Overworked teachers in southern India have a new tool for educating their students: Shiksha Copilot, an AI system that generates culturally relevant lesson plans for schools with few resources.
Researchers at Cornell and Microsoft Research India evaluated Shiksha Copilot, among more than 1,000 teachers in fifth- to 10th-grade classrooms in the state of Karnataka, India. Initially developed by Microsoft Research India, the system is named after the Sanskrit word for "learning." Teachers reported the tool, which provides curated lesson plans in English and the local language, Kannada, saved them time and stress, and allowed them to replace lecturing with more hands-on activities.
"With English, there are lots of resources available on the internet already, but that's not the case for languages like Kannada," said Deepak Varuvel Dennison, a doctoral student in the field of information science and first author of the research. "Here, we have a tool where the lesson plans have already gone through a vetting process by experienced teachers who ensure they are high quality and devoid of any errors."
Varuvel Dennison will present the study, "Shiksha Copilot: Teacher-AI Collaboration for Curating and Customizing Lesson Plans in Low-Resource Schools," at the ACM Computer-Supported Cooperative Work and Social Computing conference, held Oct. 10-14, in Salt Lake City, Utah. The paper received an Honorable Mention.
Shiksha Copilot is composed of a collection of AI models that convert textbook materials into grade-appropriate lesson plans in English and translate them into Kannada, using guardrails that ensure culturally relevant examples - such as dosas instead of pizza when talking about popular foods. A set of curators reviews the lesson plans for accuracy, and teachers further customize them for their classrooms.
The project brought together Aditya Vashistha, assistant professor of information science in the Cornell Ann S. Bowers College of Computing and Information Science, and Tanuja Ganu, director of research engineering at Microsoft Research India, around a shared interest in making AI technologies - many of which are developed primarily with Western contexts in mind - more useful and accessible to people in India and across the Global South.
"Technologies are designed for the first 1 billion people who use them, and an enormous amount of effort is required to take the technology to the next billion, and the next billion, and so on," Vashistha said. "By the point you come to the last mile, which is a place where lots of governments, nonprofits, social enterprises, think tanks and even industry is excited about, there is this big disconnect in what technology is designed and what it is being used for."
To evaluate the technology and ensure it worked correctly for its intended users, the researchers teamed up with the Sikshana Foundation, a local nonprofit working to improve government schools, with the goal of providing a tool that not only helps teachers, but also encourages long-term learning for students.
Critically, the tool had to provide lesson plans that are "well-tailored, in terms of the cultural nuances, local languages, local curriculum, and the resources which are available," Ganu said. "That was, if not more effort, at least equal effort to building the technology."
Curators reported the vast majority of the Shiksha Copilot English lessons plans were high quality. Of 7,744 plans, they made minor adjustments to 3.53% and rejected just 0.34% for technical errors. The lesson plans in Kannada were less reliable, however, with 96% needing changes. Many of these edits were due to nuances of terminology, such as substituting "mirrors" for "glasses" or replacing "chronology" with "timeline." However, teachers still deemed almost 81% of those plans "very good."
Based on interviews and surveys, teachers said Shiksha Copilot saved them an average of two hours each week and eased the burden of administrative tasks, like submitting hard copies of their lesson plans.
"Usually, whenever someone comes with something new, it just adds to our workload, but this is a burden-releasing concept," said a science teacher with 20 years of experience.
Teachers also reported that Shiksha Copilot helped them plan more hands-on activities, such as collecting local leaves with different vein patterns.
Now, the team is testing Shiksha Copilot in a larger population. This year, about 30,000 teachers are using the tool across the state of Telangana, India.
This work is part of broader efforts by Vashistha, Ganu and Varuvel Dennison to encourage the development of culturally aware AI systems that address education, healthcare, law and agriculture needs in the Global South. They recently developed a framework to guide the design and deployment of these AI technologies, which they presented at the 2026 CHI Conference on Human Factors in Computing Systems, held in April in Barcelona. Their analysis of Shiksha Copilot and seven other AI systems for social good deployed in different countries showed these projects require extensive collaboration with experts on the ground. Tailoring an application so it integrates with local systems and effectively addresses local needs can account for up to 80% of the work to develop it, they found.
"Especially in low-resource settings, there are so many different kinds of surprises we typically encounter," Ganu said.
Co-authors on the paper include Bakhtawar Ahtisham, a doctoral student in the field of information science; Kavyansh Chourasia, Nirmit Arora, Rahul Singh and Akshay Nambi, all from Microsoft Research India; and Rene Kizilcec, associate professor of information science in Cornell Bowers.
Shiksha Copilot is supported by Microsoft in collaboration with the AI Innovation and Inclusion Initiative at the International Institute of Information Technology Bangalore and the Sikshana Foundation, with evaluation supported by Global Cornell, the Cornell Global AI Initiative and Microsoft.
Patricia Waldron is a writer for the Cornell Ann S. Bowers College of Computing and Information Science.
