EduGrade AI

Inspiration

Grading code repositories and complex student assignments is often time-consuming and inconsistent for educators. Traditional automated tools rely on rigid test cases without understanding semantic code structure. We built EduGrade AI to revolutionize classroom evaluations by combining Qwen Cloud's advanced reasoning capabilities with strict educator control.

What it does

EduGrade AI operates as an autonomous Teaching Assistant ecosystem:

Context-Aware Evaluation: Deeply analyzes code files, project architecture, and logic against custom academic rubrics using Qwen Cloud models.
Granular Feedback Engine: Explains scoring rationale, pinpoints exact lines needing refactoring, and suggests modern architectural best practices.
Human-in-the-Loop (HITL) Guardrail: Keeps all AI evaluations in a "Pending Approval" state on a secure Teacher Dashboard, allowing educators to review, edit, or override scores before final release.
Autopilot Communication Agent: Automatically dispatches formatted assessment reports directly to students via email once approved by the educator.

How we built it

Multi-Agent Orchestration: Driven by CrewAI and LangChain running on Python (backend/agents.py).
Frontend & Dashboard: Built using Streamlit and FastAPI (frontend/ui.py) for real-time approval workflows.
LLM Engine: Qwen Cloud API via Alibaba Cloud DashScope for deep code reasoning.
Cloud Infrastructure: Hosted on Alibaba Cloud ECS (Elastic Compute Service).
Devin Integration: Devin AI was leveraged to refactor CrewAI agent configurations, audit asynchronous FastAPI endpoints in app.py, and construct automated testing harnesses.

Challenges we ran into

Synchronizing state between the asynchronous CrewAI multi-agent workers and Streamlit's reactive UI.
Structuring systemic grading prompts for Qwen Cloud to deliver consistent JSON schemas for the HITL dashboard gate.
Configuring low-latency SMTP background workers on Alibaba Cloud ECS to handle batch email notifications.

Accomplishments that we're proud of

Successfully deployed an end-to-end multi-agent grading pipeline on Alibaba Cloud ECS.
Implemented zero-hallucination guardrails via a seamless Human-in-the-Loop approval gate.
Combined deep code reasoning with automated student feedback loops into a single zero-friction platform.

What's next for EduGrade AI

LMS Integration: Build native plugins for Canvas, Moodle, and Google Classroom.
Plagiarism & AI Detection Matrix: Integrate vector-similarity checks to evaluate submission originality alongside quality.