Inspiration

As an educator with nearly 10 years of experience teaching middle school, I have always believed that technology should make teaching easier, more meaningful, and more impactful.

My journey into AI began with an AI in Education workshop conducted by IIT Madras, Chennai, where nearly 5,000 educators participated online and only 50 were selected for the in-person workshop. That experience became a turning point for me. I learned not only about the transformative possibilities of AI in education but also about how AI works technically, which inspired me to explore it deeply.

When I returned to school, I saw how many teachers—especially those focused on core subjects—were struggling to confidently adopt AI. I began asking myself: “How can I make AI genuinely accessible, authentic, and useful for every teacher without requiring complex prompting?” That question became the foundation for NESA EDU AI AGENT.

What it does

NESA EDU AI AGENT — Self-Learning AI Teaching Partner is designed to go beyond being a simple chatbot or prompt generator. It acts as an autonomous, self-learning educational co-pilot, helping transform curriculum goals into classroom-ready learning experiences.

Instead of asking teachers to master complex prompt engineering, NESA EDU AI AGENT is designed to understand the teacher's requirement, plan, create, validate, improve, and deliver comprehensive educational resources. Key capabilities include:

30 Specialized AI Engines: Across Plan & Teach (5E Lesson Plans, Annual Maps), Create & Engage (Interactive Canvas Apps, Visual Diagrams, Video Storyboards), Practice & Assess (3-Tier Worksheets, Bloom's HOTS Question Banks, HPC 360° Rubrics), and Inclusion (Emerging Learner & Divyangjan UDL scaffolding).
Pan-India A-to-Z Curriculum: Authentic mappings for all Indian State & National Boards (CBSE, ICSE, Tamil Nadu, Maharashtra, Karnataka) across Grades 1 to 12.
Multi-Modal Generation: Live educational diagrams, SVG vector graphics, and in-browser playable HTML5 physics simulations.
1-Click Classroom Export: Instant multi-format export to PDF printouts, Markdown (.md), Plain Text (.txt), and Text-to-Speech audio readout.

How we built it

We designed NESA EDU AI AGENT around a true agentic architecture powered by Google Gemini 3.5 & 2.5 Flash on Vertex AI and Google Agent Development Kit (ADK) principles.

Agentic Routing & Capabilities: The 30 engines are treated as specialized capabilities within an intelligent engine registry, allowing the agent to understand teacher intent and dynamically route tasks to the appropriate pedagogical workflow.
Reflexion Self-Learning Memory: Implemented an autonomous self-correction loop where teacher feedback and corrections are synthesized into pedagogical rules and stored in Google Cloud Firestore, continuously adapting to the educator's teaching style.
Full-Stack Architecture: Built with a high-performance FastAPI backend utilizing asynchronous Server-Sent Events (SSE) streaming, containerized with Docker for Google Cloud Run, and paired with a modern, accessible Next.js 14 / TypeScript frontend featuring dual dark/light themes and voice dictation.

Challenges we ran into

The biggest challenge was not simply connecting an AI model to an API. The real challenge was understanding what teachers actually face in daily classroom realities: time constraints, diverse board curricula, multi-grade learning levels, and varying digital literacy.

Another major engineering challenge was transitioning from a prompt-based mindset to an autonomous agentic mindset—enabling the system to decompose complex educational objectives, ground them against authentic Indian curriculum standards (NEP 2020 & NCF 2023 Panchpadi pedagogy), and self-correct via Reflexion memory rather than returning generic responses.

Accomplishments that we're proud of

Transformed a vision born from everyday classroom experience into an autonomous, self-learning teaching co-pilot.
Integrated 30 specialized pedagogical engines that stream 15,000+ character curriculum plans in under 3 seconds.
Replaced outdated labels with modern, respectful inclusive terminology (Emerging Learners, Accelerated Learners, Divyangjan).
Built an interactive simulation player that lets students and teachers interact with physics and science experiments directly in the workspace.
Most importantly: Ensured that teachers do not have to become AI experts to benefit from AI.

What we learned

This journey taught us that building AI for education is not merely about choosing the largest model or adding superficial features. It is about deeply understanding pedagogy, teacher cognitive load, curriculum nuance, and student diversity.

We learned that AI becomes indispensable when it eliminates administrative friction and returns precious time back to teachers. We also learned that an agentic architecture with continuous feedback, reflection, and memory is essential for building an AI that educators can genuinely trust day after day.

What's next for NESA EDU AI AGENT

Our vision is to evolve NESA EDU AI AGENT into a permanent intelligent teaching partner for Indian educators:

Regional Language Scaling: Fine-tuning regional models across all 22 scheduled Indian languages for vernacular classrooms.
National Platform Integration: Seamless interoperability with DIKSHA, SWAYAM, and State LMS platforms.
Edge AI Deployment: Enabling offline-first edge agents using Google Gemma 2 for rural, low-connectivity schools.

Ultimately, our goal is not to replace teachers, but to amplify their expertise, creativity, and impact—giving every teacher an intelligent partner so they can focus on what matters most: their students.

Built With

agent-development-kit
docker
fastapi
google-cloud-firestore
google-gemini
googlecloudrun
next.js
python
typescript
vertex-ai

Updates

Nesakumar Mohandass started this project — Aug 30, 2026 05:06 PM EDT

Leave feedback in the comments!

Log in or sign up for Devpost to join the conversation.