2026 · AI product
Crescendo
AI project management in which every model call receives the full board state, not an isolated card.
20agent tools: 5 read, 15 that can only propose a change
- My role
- Built for the Opus 4.7 Hackathon in April 2026: the full-stack app and the Board-State Awareness Engine behind it.
- Result
- Every model call sees the whole board, nothing is written until a person confirms it, and assigning work to someone at burnout risk is refused until that risk is acknowledged.
- Built with
- React 19, TypeScript, Node.js, Fastify +2 more
Problem
AI assistants in project tools answer about one card at a time. They don't see who is overloaded, what is blocked, or which dependency is about to slip.
My role
Built for the Opus 4.7 Hackathon in April 2026: the full-stack app and the Board-State Awareness Engine behind it.
What I built
- Board-State Awareness Engine grounds every response in complete project context
- Propose-then-confirm write path enforced in code, not in a prompt
- RBAC with 5 roles, JWT sessions, and audit logging
- Kanban, Gantt timeline, list, files, activity, and an agent pane
Outcome
Every model call sees the whole board, nothing is written until a person confirms it, and assigning work to someone at burnout risk is refused until that risk is acknowledged.
