EduGrade AI — Smart Grading & Feedback Ecosystem

💡 Inspiration

Traditional grading is often a bottleneck in education, leaving teachers overwhelmed with mounting piles of assignments and students waiting days for feedback. We wanted to create a solution that bridges this gap. The inspiration behind EduGrade AI was to build an autonomous, dual-agent system that not only automates the grading process with high accuracy using Qwen Cloud but also ensures a professional, human-approved workflow (Human-in-the-Loop) to maintain trust, precision, and transparency in academic evaluations.

🚀 What it does

EduGrade AI is a comprehensive, production-ready smart grading and feedback ecosystem designed specifically for modern educators:

• Autonomous Grading Pipeline: Students can submit assignments (essays, code, or written answers) along with highly customized, multi-criteria evaluation rubrics.
• Dual-Agent Architecture: Powered by the Qwen Cloud LLM, the Grading Agent rigorously scores submissions based on strict parameters, while the Communicator Agent automatically drafts personalized, professional feedback and email reports.
• HITL (Human-in-the-Loop) Dashboard: Grading is never fully black-box. Educators have a dedicated Review Queue where they can audit AI-generated grades, override scores if necessary, and officially approve assessments before final dispatch.
• Command Center: A real-time analytical dashboard displaying total submissions, pending review counts, approved records, and average final scores.

🛠️ How we built it

We engineered EduGrade AI using a modern full-stack architecture optimized for fast inference and deployment:

• Frontend: A highly responsive and crisp dashboard built using React and Vite, featuring clean UI components, modular views, and dynamic status badges.
• Backend: A robust Node.js & Express.js API server that serves as the orchestrator for data flows, prompt pipelines, and state transitions.
• Database: An integrated SQLite / In-Memory queue architecture to securely manage pending submissions, rubrics, and final approved evaluation reports.
• AI Core: Integrated with Alibaba Cloud DashScope (Qwen Cloud) via Replit's secure AI gateway to process complex prompt chains, ensuring high-fidelity structured JSON outputs for grades and feedback.

🚧 Challenges we ran into

One of our primary hurdles was optimizing the latency between executing multi-agent reasoning (grading + communication drafts) and rendering real-time state updates in the UI. Initially, chaining consecutive LLM queries created a bottleneck. We solved this by structuring asynchronous endpoints in Express and decoupling the grading logic from the state management using an efficient database-driven queue. This made the entire interface snappy and seamless across both desktop and mobile viewports.

🏆 Accomplishments that we're proud of

• Successful Migration: Transitioning from a basic script concept into a fully optimized, production-ready React + Express full-stack web application.
• Foolproof HITL Design: Successfully implementing a true Human-in-the-Loop (HITL) validation layer that bridges the efficiency of AI with the irreplaceable authority of human educators.
• Flawless Live Deployment: Spinning up the entire ecosystem live on Replit with an interactive, pre-populated "Command Center" dashboard functioning flawlessly out of the box.

📚 What we learned

We gained profound insights into prompt engineering, chaining, and structured output formatting using the Qwen Cloud models. We discovered how reliable Qwen is at adhering to strict, deterministic evaluation rubrics. Additionally, we mastered designing multi-agent state architectures where one agent's analytical evaluation seamlessly forms the baseline context for a second agent's communication model.

🔮 What's next for EduGrade AI

Our roadmap for scaling EduGrade AI includes:

1. Multi-File & PDF Parsing: Allowing students to upload scanned exam scripts, multi-page PDFs, or entire project directories natively.
2. Plagiarism & AI Detection Integration: Introducing automated cross-referencing against web sources and AI-generated content checking directly inside the ingestion pipeline.
3. LMS Ecosystem Connectors: Developing dedicated webhooks and plugins to seamlessly sync grades with mainstream Learning Management Systems like Moodle, Canvas, and Google Classroom.

Built With

• ai
• alibabacloud
• automation
• css3
• express.js
• html5
• javascript
• llm
• node.js
• qwen
• react
• replite
• sqlite
• tailwindcss
• vite