Vibe Coding the MIT Course Catalog
18 days ago
- #MIT
- #course discovery
- #AI-assisted development
- Transition from Microsoft to MIT's Media Arts and Sciences program highlighted challenges in course selection, especially discovering 'unknown unknowns'.
- Existing tools like MIT's official course catalog and Hydrant are outdated or inefficient, lacking features for simultaneous course comparison and AI assistance.
- Courseek was developed to address these issues, leveraging AI-assisted development and testing the limits of LLM capabilities in course discovery.
- The project involved scraping MIT's course catalog, benchmarking existing tools, and iterative prototyping to improve search functionality and user interface.
- Challenges included handling special characters in search terms and optimizing the search algorithm for performance and accuracy.
- Final architecture features a three-stage pipeline: matching, scoring, and highlighting, with a focus on exact and prefix matches for better search results.
- Automation pipeline ensures course data is updated daily via GitHub Actions, keeping the tool current without server infrastructure.
- Future improvements may include a more robust full-text search engine and integration with Model Context Protocol for chat-based queries.