Skip to content

Canvas AI Agent

Presented as HackWashU 2025 Project, github.

Live Demo: https://markso.ng/demo/canvas

The Canvas AI Agent is an ReAct based OpenAI based learning assistant that integrates directly with the Canvas LMS, combining natural-language interaction with live academic data retrieval. Developed during HackWashU 2025 as part of the AI track, our team built a fully functional prototype that connects OpenAI models with university course management systems through a secure, scalable backend.

Key Contributions

  • System Design: Architected an end-to-end AI-assisted learning system integrating Canvas APIs, OpenAI GPT-5, and Vector Stores for contextual retrieval and reasoning.
  • Backend Engineering: Implemented an asynchronous FastAPI backend supporting concurrent multi-user sessions and live WebSocket (wss://) communication.
  • Frontend Development: Built a responsive React + TypeScript interface using Ant Design, enabling real-time chat interaction and visual feedback.
  • Knowledge Integration: Developed 26 student-permissioned tools (22 Canvas tools + 4 Vector Store tools) to retrieve assignments, grades, files, and course modules with semantic search.
  • AI Engine: Integrated Azure OpenAI’s GPT-5 with reasoning and query execution memory based on SKYWORK AI architecture.
  • Performance & Accessibility: Reduced navigation time in Canvas; improved accessibility for students with disabilities and multilingual needs.
  • Impact: Designed for scalability across institutions; the system architecture can support thousands of concurrent users.

Note due to limited computational resources, price of the API key, and security reasons, the live demo is limited to a single user at a time and ten minutes per session. The demo is also using gpt-4o due to the same reasons. The full version of the project is using gpt-5 and can be deployed on your own server.