Skip to content
← Home

Lazysheep Coursemate

Independent Product Lead · Feb 2025 – Present

Built an LLM course-selection agent for scattered registrar information.

Reached 2,000+ uses and 400 campus users.

View project launch (Chinese)
01 / 03Look up a courseAsk with a course code or name, and read the credits, language, and whether it is offered.

Look up a course

Lazysheep reply introducing ECE3080 with credits, language, and whether it is offered

Press Esc or select Close to return to the project.

Problem

Registrar facts were scattered, so choosing a course meant hunting across pages.

Decision

A RAG agent over 10,000 official course records, with a Plan–Reflect–Question loop when the first recall is not enough.

Shipped

  • Engineered the knowledge base from 10,000+ official course records for search, comparison, and recommendations.
  • Built a RAG stack with intent recognition and multi-path recall across syllabus, grading, and other sources.
  • Architected a Plan–Reflect–Question workflow on ReAct, lifting context recall to 88%.
  • Ran campus promotion and community ops to grow adoption.

RAG · Agents