Deepak Varuvel Dennison (center), a Ph.D. student in information science at Cornell University, and Rijutha Raj (right) of the Sikshana Foundation speak with a middle school teacher about their experience using Shiksha Copilot.
ITHACA, N.Y., Oct. 8, 2026 | Researchers at Cornell University and Microsoft Research India have built an AI tool that turns textbook material into lesson plans using local examples. It was tested with more than 1,000 teachers in grades 5 to 10 in Karnataka, India, Cornell reports. Teachers estimated the tool, called Shiksha Copilot, saves them about two hours a week.
Microsoft Research India developed the system for under-resourced schools. It generates grade-appropriate lesson plans in English and translates them into Kannada, the local language. Guardrails steer examples toward local context, such as dosas rather than pizza. Experienced curators check each plan for accuracy, and teachers then customize it for their own classrooms. The work was done with the Sikshana Foundation, a nonprofit that works to improve government schools.
What the Study Found
The research team evaluated the tool's output and teachers' experience. Of 7,744 English plans, curators made minor adjustments to 3.53% and rejected 0.34% for technical errors. Kannada plans needed far more work: 96% required changes, mostly for terminology, such as using "mirrors" instead of "glasses" or "timeline" instead of "chronology." Teachers still rated almost 81% of the Kannada plans "very good."
Beyond time savings, teachers said the tool reduced administrative work, such as submitting hard-copy lesson plans. Some used it to move from lecturing toward hands-on activities, such as collecting local leaves with different vein patterns. "Here, we have a tool where the lesson plans have already gone through a vetting process by experienced teachers," said one of the researchers, according to Cornell.
The results are teacher-reported and based on plan quality, not on student test scores. Cornell's account does not report long-term learning outcomes, and the larger rollout is still under way.
Why Local Adaptation Is the Hard Part
Kannada has far fewer online educational resources than English, which is why the researchers built in specific language support. Many AI tools are designed mainly for Western contexts, which leaves a gap for users in India and the broader Global South.
The Cornell team found that local adaptation takes serious effort. A paper presented at the CHI 2026 conference in Barcelona analyzed Shiksha Copilot and seven other AI social-good projects in different countries. It found these systems depend heavily on collaboration with local experts, and that tailoring a system to local needs and systems can account for up to 80% of development work.
The study, "Shiksha Copilot: Teacher-AI Collaboration for Curating and Customizing Lesson Plans in Low-Resource Schools," earned an Honorable Mention. It will be presented at the ACM Computer-Supported Cooperative Work and Social Computing (CSCW) conference in Salt Lake City, Oct. 10 to 14. The first author is Deepak Varuvel Dennison, a doctoral student in information science at Cornell, and co-authors include Aditya Vashistha, Rene Kizilcec and Bakhtawar Ahtisham at Cornell and Tanuja Ganu and colleagues at Microsoft Research India. Cornell says about 30,000 teachers are using the tool this year across Telangana, India.
Frequently Asked Questions
What is Shiksha Copilot?
Shiksha Copilot is an AI system that creates culturally relevant lesson plans for under-resourced schools in southern India. It turns textbook material into grade-appropriate plans in English, translates them into Kannada, and lets curators and teachers review and customize them.
How much time does it save teachers?
Teachers in the study estimated it saves about two hours a week. That figure is self-reported, and the researchers have not published long-term student learning results.