
Maritime learners need study support that works even at sea without reliable internet. This project used IBM Bob IDE to reproduce, fix, and verify a real defect in an existing maritime Learning Management System (LMS): the built-in AI assistant was firing token requests and creating spurious sync items even when the learner was offline, blocking the offline learning flow. IBM Bob drove the full developer workflow — reproducing the defect in real Chrome, writing and fixing Angular integration tests (101 passing, exit 0), adding an offline-safe sidebar with clear provider state indicators, and verifying layout on desktop and mobile. New capabilities added within the hackathon: • Offline drafting mode — no requests queued or replayed when connectivity returns • Explicit opt-in local inference via Ollama on the same device (no automatic probe, no LMS token attached) • Optional personal ChatGPT provider via device-code OAuth — disabled by default, tokens isolated per user • Selected lesson text copied to assistant only on explicit user action All changes were verified with real Chrome (153.0.8010.53), 101 targeted Angular tests, Docker Compose validation, and the project harness. Live ChatGPT inference remains experimental and clearly labeled. The LMS baseline and PWA infrastructure are preexisting; the offline fix, local provider, and test coverage are the hackathon contribution.
27 Sep 2026