Canvas Coding Assistant

Production LLM coding assistant for Canvas LMS — Qwen2.5-Coder-7B on GPU, serving 100+ active users at the University of Houston.

Overview

Canvas Coding Assistant is an AI-powered coding helper integrated into the University of Houston's Canvas LMS workflow. A Chrome extension captures code from OneCompiler assignments, sends it to a FastAPI backend, and returns a single actionable suggestion generated by Qwen2.5-Coder-7B-Instruct running on GPU via SGLang.

Built as part of my Research Assistant role at UH, the system is in active production use across introductory programming courses.

Impact

• 100+ active users in production across UH programming courses
• GPU-hosted inference via SGLang on Qwen2.5-Coder-7B-Instruct
• Chrome extension embedded directly in the student coding workflow (OneCompiler + Canvas)
• Structured request logging — prompt size, platform, language, and latency tracked per request

Architecture

flowchart LR
    Ext["Chrome Extension"] -->|"POST /suggest"| API["FastAPI Backend"]
    API -->|"chat completions"| SGL["SGLang Server"]
    SGL --> Model["Qwen2.5-Coder-7B"]
    Model --> SGL
    SGL --> API
    API --> Ext

Loading

sequenceDiagram
    participant Student
    participant Extension
    participant FastAPI
    participant SGLang
    participant Model as Qwen GPU

    Student->>Extension: Analyze code
    Extension->>FastAPI: POST /suggest
    FastAPI->>SGLang: Chat completion
    SGLang->>Model: Inference
    Model-->>SGLang: Generated text
    SGLang-->>FastAPI: JSON response
    FastAPI-->>Extension: Suggestion
    Extension-->>Student: Display result

Loading

Tech Stack

Layer	Technology
Model	Qwen2.5-Coder-7B-Instruct
Inference	SGLang (OpenAI-compatible API)
Backend	FastAPI, Python 3.10+
Client	Chrome Extension (Manifest V3)
Transport	REST

Quick Start

1. Start SGLang

python -m sglang.launch_server \
  --model-path Qwen/Qwen2.5-Coder-7B-Instruct \
  --host 127.0.0.1 \
  --port 30000

2. Configure the backend

cd backend
python -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp ../.env.example ../.env   # edit MODEL_ID / SGLANG_BASE_URL if needed

3. Start FastAPI

uvicorn main:app --host 127.0.0.1 --port 8000 --reload

Verify:

curl http://127.0.0.1:8000/health

4. Load the Chrome extension

1. Open chrome://extensions
2. Enable Developer mode
3. Click Load unpacked and select the extension/ directory
4. Navigate to a OneCompiler page and use the extension popup to analyze code

Project Structure

canvas-coding-assistant/
├── backend/
│   ├── main.py              # FastAPI app + /suggest endpoint
│   └── requirements.txt
├── extension/
│   ├── manifest.json        # Manifest V3
│   ├── content.js           # Code extraction from OneCompiler
│   ├── popup.html/js        # Extension popup UI
│   └── inject.js
├── docs/
│   └── architecture.svg     # Architecture diagram
├── .env.example
├── LICENSE
└── README.md

Configuration

Variable	Default	Description
SGLANG_BASE_URL	http://127.0.0.1:30000	SGLang OpenAI-compatible API base URL
MODEL_ID	Qwen/Qwen2.5-Coder-7B-Instruct	Hugging Face model identifier
CORS_ORIGINS	*	Comma-separated allowed origins (restrict in production)

Security Notes

• Development only: default CORS_ORIGINS=* allows any origin — restrict to your extension origin before production deployment.
• No secrets in repo: use .env for configuration; never commit credentials or internal paths.
• Production hardening: add HTTPS termination, authentication on /suggest, and rate limiting before public exposure.

Author

Jyotiradityasinh Chauhan

Portfolio · LinkedIn · GitHub