A Kanban board that survives a refresh.
Task management with authentication, a Kanban board, assignment, search and filtering, backed by a real database rather than local state.
01 / The problem
The brief asked for task management with authentication, a Kanban board, assignment to team members, search and filtering.
The real difficulty was building something that genuinely stores and manages data, rather than a Kanban that only looks like it works because everything lives in local state. Tasks had to still be there after a refresh, and changes to status, assignee, deadline and user had to persist consistently on the backend.
02 / Decisions
A real database with migrations, not an in-memory mock
I used PostgreSQL with an actual schema and migrations instead of holding data in memory or standing up a mock backend. The point was persistence, and a database setup that reproduces cleanly when someone else runs it on their own machine.
The AI assistant classifies intent and nothing else
Gemini reads the user's question and picks one of four predefined intents. After that the backend runs functions and queries I wrote myself. I rejected letting the model generate SQL directly: more flexible, but not worth the risk for a simple read-only feature. I would rather have an assistant that can do less and a data path I can fully account for.
03 / The outcome
- JWT authentication with a seeded demo account.
- A three-column Kanban board with status and assignee changes.
- Search by title, plus filtering by status, assignee and deadline.
- A My Tasks view and a task detail drawer.
- Interactive Swagger documentation.
- Docker Compose for PostgreSQL, Alembic migrations, and a reproducible seed script.
- Setup documentation for macOS and Windows, so a reviewer can run it from an empty environment.