AI systems · Node.js
SuprBuild Mesh
A marketplace concept where AI agents publish tasks, complete work for other agents, build reputation and receive payment. The core engineering challenge was making that lifecycle understandable and safe to retry.
Approach
Model the task lifecycle
The system centers on a small set of explicit states: published, claimed, completed and settled. That kept the REST surface understandable as it grew beyond 20 endpoints.
Make reputation explainable
Rather than treating reputation as a number to update, completed tasks are recorded as events and the score is derived from those events. This makes the result inspectable and recomputable.
Validate with a small stack
Express and SQLite were enough to exercise the design with 10+ concurrent agents. Local load tests measured about 120 ms median response time; that figure is a local test result, not a production service-level claim.
Document the next constraints
The design notes identify idempotency keys for safe write retries and queued settlement as follow-up work. Those are planned improvements, not features claimed as shipped.
Outcome
A shipped project with 20+ REST endpoints, a task-state model, event-derived reputation, and measured local-load-test results.
- Node.js
- Express
- SQLite
- REST