Skip to content
Second Place

Sahayak AI

Vertical Copilot for Education - AI for the Next Generation of Indian Teacher

Core trick

OCRs textbook page images into in-context prompts that generate curriculum-aligned quizzes, and an MCP server renders Manim animations.

Tags

Built with

claude-codeclaude-sonnet-4.5-apifastapifirebase-firestoreflutter-(dart)google-cloudlangchainmcppythonrag

Description

Demo Video https://dub.sh/SahayakAnthropicDemo (not in the video field because we had a drive link) Inspiration We wanted to leverage AI to empower these educators with tools for lesson planning, assessment generation, and classroom management without requiring technical expertise. We have built Sahayak a vertical copilot agent for the next generation of Indian education What it does Sahayak AI generates curriculum-aligned quiz questions from textbook images using OCR and In-context learning, creates animated educational videos with Manim video generation via an MCP, and automates attendance tracking. All content is NCERT-aligned and available in multiple Indian languages. How we built it FastAPI backend powered by Claude Sonnet 4.5, integrated with Google Cloud (Firestore vector store, Cloud Storage). LangChain for RAG pipelines, PostgreSQL for data, and Flutter frontend for teacher-friendly interface. Challenges we ran into Chunking NCERT PDFs while preserving context across chapters and topics, handling diverse formatting in Indian textbooks, and mapping vector embeddings to structured curriculum data. Also, ensuring the MCP server could auto-correct code errors in the Manim Animation code proved to be a significant technical hurdle. Accomplishments that we're proud of Built a production-ready multi-agent system handling complex educational workflows end-to-end. Our quiz pipeline generates NCERT-aligned questions using in-context learning, and Manim animation engine integration creates professional animations from simple teacher prompts that make explaining complex topics easy. What we learned Building robust multi-agent AI systems with Claude 4, implementing effective RAG with vector databases, and designing teacher-centric UX for non-technical users in resource-constrained environments. What's next for Sahayak AI Expand to more regional languages, add lesson planning agents, integrate real-time classroom analytics, and develop offline-first capabilities for schools with limited internet connectivity.

The linked repo returns 404, so it was deleted or made private.

Team