Smart Grade AI

Recognition

Microsoft Imagine Cup 2025 — Global Semifinalist (only Asian team to qualify from 20,000+ submissions)

Microsoft for Startups — $25,000 in Azure credits awarded

Accepted for institutional integration into the University of Management and Technology (UMT) LMS

Problem

Manual grading is slow, inconsistent across markers, and produces shallow feedback. Teachers also lack tooling to detect class-wide weakness patterns or guide individual students through their actual gaps.

Solution

Smart Grade AI is an end-to-end LMS that handles assessments, grading, feedback, and personalised mentoring in one flow. Teachers configure rubrics; students submit PDFs or images; the AI grades against the rubric and produces structured per-question feedback. Every graded assessment is embedded into a per-student vector store that powers a personal AI Mentor chatbot, while an AI Teacher Assistant aggregates class-wide signals for the instructor.

Key technical features

Rubric-based AI grading engine: Handles text answers, coding submissions, and diagram/vision-heavy papers. Outputs question-wise marks, detailed feedback, and structured JSON used downstream by analytics and RAG systems.

Per-student AI Mentor (RAG): Every graded assessment is embedded into a student-scoped vector store. The chatbot retrieves relevant past performance signals and produces grounded, personalized advice — surfacing weakness patterns, recommending study plans, and answering performance questions.

AI Teacher Assistant: Delivers class-wide insights ("which questions caused the most failures?"), struggling-student detection, weekly reporting, and remedial practice suggestions.

End-to-end pipeline: Mistral OCR for document/image text extraction → LangGraph grading workflow → Azure OpenAI evaluation → PDF and Excel report generation → vector store update.

Stack

React 18 with TypeScript, Vite, Tailwind CSS, shadcn/ui, Framer Motion, Recharts. Clerk authentication with role-based access. Node.js / Express for the LMS core, FastAPI with LangGraph for the AI service. MongoDB / CosmosDB. Azure OpenAI (GPT-4 class models), Mistral OCR. Cloudinary, Docker, Azure Container Apps.