IRIS
IRIS takes the manual work out of sprint planning. Drop in a product doc, and AI breaks it into a structured backlog — grouped features, prioritized tasks, and clear dependencies — synced directly to Redmine.
Case Study
My Role
Solo build — pipeline architecture, backend, frontend, and the Redmine integration.
AI Pipeline — core engineering
Designed a phased LLM orchestrator: an Analyze phase reads the PRD and detects platforms and suggested tech stacks; a Discover phase sections the document and groups related content into ranked "plannings" with typed dependencies (prerequisite vs. loosely related); an Extract phase generates features with confidence scores and dependency-aware implementation todos, tagged by team (frontend/backend/infra). Each phase pauses for a checkpoint, letting the AI ask clarifying questions before continuing — turning a one-shot generation into a reviewable, resumable process.
Data Model
An 11-table database tracks projects, plannings, features, and todos alongside pipeline run state, checkpoint Q&A history, and AI-suggested impact changes — enough structure to support pause/resume pipelines and full audit history per project.
Frontend
A project view (PRD upload, pipeline status, architecture/stack review panels), a feature view with a planning/feature hierarchy and status-toggleable todos, a graph view visualizing the knowledge graph of features and entities, and a chat panel for brainstorming directly against the project context.
Redmine Integration
Built a bidirectional sync layer — features, todos, and plannings map to typed Redmine tickets, with status mapping and a ticket-adapter pattern designed to support other issue trackers beyond Redmine.