Problem
Developers need evidence that a coding agent’s completion claim matches what actually happened in its trace.
TracePilot AI
A multi-agent platform that audits coding-agent execution traces using blind outcome verification and stateful reliability tracking.
Architecture
A React frontend submits traces to a Spring Boot orchestrator, which persists and dispatches jobs through RabbitMQ to a FastAPI/CrewAI worker. PostgreSQL stores audits and Jaeger receives distributed traces.
Role
Architected the polyglot microservices platform, the asynchronous audit pipeline, reliability reports, observability, and the React dashboard.
Evidence
The project includes blind-verifier prompt isolation tests, asynchronous multi-agent audits, stateful reliability reports, distributed tracing, and shareable audit output.