Agentic LMS Automation System

Agentic browser automation for LMS workflows (Moodle and Canvas).

An agentic browser automation system that autonomously manages LMS workflows (Moodle and Canvas). The system leverages LLMs to navigate complex web interfaces, identifying and completing student tasks without manual intervention.

Technical Stack

  • Agent Framework: Microsoft AutoGen
  • Automation: Playwright
  • Intelligence: OpenAI GPT-4o (also supports
  • Backend & Ops: Docker
  • Observability: LangSmith

Core Features

  • Autonomous Navigation: Orchestrated an end-to-end “discovery-to-submission” workflow, replacing rule-based scripts with an adaptive, LLM-driven architecture.

  • Intelligent Browser Logic: Developed an HTML-to-Action parser using ARIA snapshots and GPT-4o to interpret complex DOM structures and execute Playwright actions.

  • Agentic Orchestration: Designed a multi-agent system via AutoGen to modularize secure authentication, assignment parsing, and content generation.

  • Observability: Integrated LangSmith to trace reasoning chains, monitor tool usage, and optimize cost/latency metrics.

Current Status: Instructor Agent integration is currently in development to automate grading and feedback loops.