The problem
In many teams, documentation lives in Confluence and work lives in Jira, and moving from one to the other is manual. It's slow and uneven, and it's exactly where you'd want AI to help, provided you know where it can be trusted.
The approach
- Reading the documentationConfluence pages are chunked and indexed in a vector database (Qdrant).
- Generating ticketsEpics and user stories with acceptance criteria, preserving tables, images and links.
- “AI-ready” verdictEach story is rated as something an AI can handle, or “needs review” with the reason.
- Code analysisA dedicated service indexes the existing codebase to give agents the right context.
The platform is the foundation of Heliocraft Agent OS, which calls it as a service.
Stack
- Java 21
- Spring Boot
- React + Vite
- Qdrant
- Claude API
- Ollama
- Confluence API
- Jira API
Results
- 475 tickets generated from real Confluence documentation.
- Structured generation runs on a local model, the “ai-ready” judgement on a stronger one: the right model in the right place, with costs under control.
- On-demand re-assessment of a story once the need is clarified.
The vision: a control plane that intercepts, evaluates, routes and audits AI interactions across the company.
Next project : Immo AI