What it does
- Users register and manage a personal wardrobe through LINE chat
- The bot fetches real-time weather from OpenWeatherMap API
- Recommends outfit combinations appropriate for the weather and occasion
- 132 passing tests, 100% code coverage
Architecture
Clean architecture in 3 layers: Presentation (LINE webhook handler), Application (recommendation logic), Domain (wardrobe + weather models).
Key Technical Decision
Kept the LLM out of the recommendation loop. The suggestion algorithm is deterministic and rule-based, which makes it testable and predictable. Adding an LLM would require mocking the external call in tests and introduces non-determinism.
Lessons Learned
- 100% coverage is achievable but not always worth the effort — the last 5% of coverage often tests framework glue code, not business logic.
- LINE's webhook verification (HMAC-SHA256) needs to be handled before any business logic runs.