%20(1620%20x%201080%20px)%20(744%20x%20400%20px).jpg)
The government aims to have every Indian school use AI, which is ambitious. However, before introducing new policies, we should first strengthen basic infrastructure.
UDISE+ data shows that only 47% of rural schools have working computers, and just 34% of villages have 4G internet. As city schools adopt cloud platforms, the gap widens between urban and rural schools, a gap that policy changes alone cannot close.
The real question is not whether to use AI in education, but which groups in India will truly benefit from it.
The Connectivity Paradox: Building an Intranet Where the Internet Doesn’t Reach
In rural schools, internet reliability is only 34%, but cloud-based AI tools require a stable connection. (Kon, 2024)
The Crescent solution is to use AI technology that processes information locally on each school's own devices, known as local-first AI, combined with edge computing, which means running AI directly on-site rather than requiring internet access to central data centres.
The Crescent Box is a device that turns a basic or older computer (such as a Mini-PC or desktop) into a local AI server (a central computer that provides AI services in the school). It runs Ubuntu, a free operating system. It uses smaller language models such as Llama-3-8B and Gemma-2-9B. These models are computer programs trained to understand and generate text in Hindi and Bhojpuri. The Crescent Box creates a local Wi-Fi hotspot, enabling teachers to access and use the AI-Saathi dashboard (a user-friendly interface) offline, even without an internet connection.
To update the system, an engineering student visits weekly with a USB drive containing the latest AI models from the college and installs them on-site. This approach uses people to overcome poor internet connections.
Project Phoenix: Reviving 47% of Schools That Have “Dead” Computers
Many school computers are unusable because Windows 10 and 11 do not run on older machines. (Singh, 2025)
The solution is to repair these computers by installing Linux.
Engineering students remove Windows (a common commercial operating system) from these computers and install Xubuntu or Ubuntu—both are lightweight, open-source operating systems that use less memory and run smoothly on older computers, even those with only 2GB of RAM and basic Pentium processors. This brings the computer lab back to working.
Schools save money on new hardware by paying a small annual maintenance fee to the engineering college. This converts a one-time, high-cost expense into a smaller, ongoing expense.
Asynchronous Intelligence: When Power Fails, Paper Delivers
Since power cuts are common, real-time AI tutoring is not always reliable.
A better approach is to use AI to prepare materials and then print them.
When power is available, teachers use the Crescent Box to create materials. With a single prompt, they can generate a custom 10-question math worksheet with diagrams, ready to print.
Students receive high-quality, AI-generated learning materials on paper. This allows lessons to continue even without electricity and helps improve teaching quality.
The Human Middleware: Why Training Teachers Is the Wrong Strategy
Only 33% of rural teachers feel prepared to use technology in their teaching. (Low Digital Readiness: A UNESCO-backed study reports that only ~33% of Indian teachers feel comfortable using digital tools for instruction, 2025) It is unrealistic to expect experienced teachers to become experts in prompt engineering.
Rather than requiring teachers to learn new technology, the solution is to provide them with technical assistants.
Instead of using complex login systems, the teacher receives assistance from an engineering student intern. The teacher sends a voice message to the Crescent WhatsApp bot (an automated helper), and either the student or the bot uses the AI tools. The finished lesson plan is sent back as a PDF, easy to read and print.
This method removes digital barriers and saves teachers 80% of their time. (Dennison et al., 2025)
Initially, the tool is used for administrative tasks like creating exam papers and writing report cards. For example, teachers who used to spend hours on test questions can now finish in minutes and reclaim that time.
Teachers are more likely to support AI when it eases administrative work, rather than altering their teaching style.
Incentives That Matter: Status, Salary, and Recognition
National policy requires schools to adopt these changes, but it does not provide personal incentives.
The solution is to offer teachers certifications that help them earn more and receive local recognition.
The Crescent Certified Educator badge is an official certificate that private school teachers can show to parents. This helps schools justify their fees and allows teachers to request higher pay. (Mendonsa, 2025)
For government school teachers, engineering colleges help nominate them for the District Innovation Awards. In government jobs, recognition is often the fastest path to promotion. (Mahapatra, 2024)
Decentralising Excellence: The Local Engineering College as IIT
IIT Delhi is a leading engineering school, but it cannot meaningfully support a rural school in Gorakhpur because their circumstances are very different, not just due to distance.
The solution is to create a network of local support centres.
Each Tier-2 engineering college (mid-level regional engineering schools) works with five to ten nearby schools as part of their community service. A "Crescent Squad" of two engineering students provides technical help to each school. The AI server is kept nearby, and support is provided in the local language by trusted community members.
Breaking the Data Invisibility Trap
AI models trained on urban and Western internet data do not align with the needs of students in rural areas like Gorakhpur. Because rural students create few digital records, algorithms are trained on urban data and may overlook rural needs.
The solution is to collect real data directly from local schools.
Engineering students gather over 5,000 handwritten exam papers from local schools to fine-tune open-source models based on UP Board standards. (Automated Test Generation and Marking Using LLMs, 2023) This ensures the AI uses examples that are relevant to the local context.
Standard Hindi may be challenging for speakers of Bhojpuri, Maithili, and Awadhi, which are local languages. To address this, engineering students record audio samples from elders and children during voice donation drives. These recordings are used to create language adapters for OpenAI’s Whisper (a tool for understanding spoken language) and similar open-source speech-to-text programs. As a result, Crescent AI can better understand questions asked in local dialects, making it more useful than generic solutions.
The Robin Hood Pricing Model
Startups often spend significant amounts on customer acquisition, resulting in annual fees of over ₹20,000 for students. (Pandey, 2022)
The solution is to use schools and colleges to reduce costs.
The engineering college charges standard rates to business clients and uses that revenue to help fund rural school infrastructure. There is no sales team or marketing budget; the college supports the school as part of its mission.
This approach enables premium AI delivery at ₹100 per student per year, or free where necessary. (Smart Classrooms on a Budget, 2026)
The Sovereign AI Promise
Connectivity | High-speed fibre, cloud-dependent | Local intranet, USB sync |
Hardware | MacBooks, i7 workstations | Revived Pentium/i3 on Linux |
Power | 24/7 supply required | Battery-ready, paper-first output |
Training Data | Urban, CBSE/ICSE bias | Hyper-local, UP Board, dialects |
Language | Standard Hindi/English | Bhojpuri, Awadhi, Maithili |
Teacher Role | Replaced by an AI tutor | Empowered by an AI assistant |
Cost | ₹2,000/student/month | ₹5,000/school/month (shared) |
Support | Remote call centre | Local engineering student |
The Crescent Ecosystem works without 5G or new hardware everywhere. Rural teachers need not be data experts, and AI benefits students without digital records.
We build robust AI solutions that perform well even with current infrastructure limitations.
This approach aligns with the current situation in India. It leverages edge computing on recycled hardware, supports local dialects through community engagement, and enables engineering students to deliver updates directly to villages.
This is what Sovereign AI means for the Global South: AI that supports development, even in areas with limited infrastructure.
Bharat’s opportunity is now. With Crescent, let’s act and close the digital divide today.
References:
- (January 13, 2025). Functional Computers in 57% of Schools, Internet in 54%: UDISE+ Report. Edconnex. https://edconnex.org/Functional-Computers-in-57-of-Schools%2C-Internet-in-54%3A-UDISE-Report
- Kon, S. S. (November 29, 2024). Rural Schools Lag 29% Behind Urban in Internet Access: Education Ministry Data. MediaNama. https://www.medianama.com/2024/11/223-29-per-cent-disparity-rural-urban-schools-internet-access-education-ministry-data/
- (n.d.). Red Hat AI 3 Validated Models. https://docs.redhat.com/en/documentation/red_hat_ai/3/pdf/validated_models/Red_Hat_AI-3-Validated_models-en-US.pdf
- Singh, R. (September 27, 2025). Windows 10 nears end: Indian PCs risk security issues without upgrade. Business Standard. https://www.business-standard.com/technology/tech-news/windows-10-support-end-2025-india-pcs-upgrade-security-risks-windows-11-125092700153_1.html
- Canonical. (November 10, 2010). AMTRON delivers 28,000 Ubuntu-based PCs to students in Assam. Ubuntu. https://ubuntu.com/blog/2010/11/10/amtron-delivers-28000-ubuntu-based-pcs-to-students-in-assam/
- (June 1, 2025). Low Digital Readiness: A UNESCO-backed study reports that only ~33% of Indian teachers feel comfortable using digital tools for instruction.. UNESCO-backed study. https://www.extensionjournal.com/article/view/2565/8-9-150
- Dennison, D. V., Ahtisham, B., Chourasia, K., Arora, N., Singh, R., Kizilcec, R. F., Nambi, A., Ganu, T. & Vashistha, A. (2025). Teacher-AI Collaboration for Curating and Customizing Lesson Plans in Low-Resource Schools. arXiv preprint. https://doi.org/10.48550/arXiv.2507.00456
- Mendonsa, K. (March 14, 2025). Private schools raise fee by 20%, teachers’ salary cited. The Times of India. https://timesofindia.indiatimes.com/city/mangaluru/private-schools-raise-fee-by-20-teachers-salary-cited/articleshow/119019454.cms
- Mahapatra, D. (May 31, 2024). SC: Promotion not a right, no criteria stated in Constitution. The Times of India. https://timesofindia.indiatimes.com/india/sc-promotion-not-a-right-no-criteria-stated-in-constitution/articleshow/110574003.cms
- (n.d.). Guidelines for Community Engagements in Higher Educational Institutions. https://coemces.manipal.edu/CER_ResourceSheets/Revised-Final-Guidelines-Community-Engagements-January-2023.pdf
- (2023). Automated Test Generation and Marking Using LLMs. MDPI 14(14). https://doi.org/10.3390/electronics14142835
- Anand, A. (September 14, 2022). 'Hindi belt' languages must thrive like Hindi: Bhojpuri, Maithili experts. India Today. https://www.indiatoday.in/india/story/hindi-belt-languages-must-thrive-bhojpuri-maithili-experts-2000315-2022-09-14
- Pandey, P. (June 1, 2022). Inside India’s Edtech Bust. Outlook Business. https://www.outlookbusiness.com/magazine/business/story/inside-indias-edtech-bust-6571
- (2026). Smart Classrooms on a Budget. Sampark Foundation. https://samparkfoundation.org/2026/01/13/building-smart-classrooms-with-low-cost-digital-tools-in-rural-india/
%20(1620%20x%201080%20px)%20(744%20x%20400%20px).jpg)