Status — Alpha / Implementation Phase The project scaffolding is complete, and the core backend logic for the vanilla agent workflow is currently under development.
An AI‑powered platform that iteratively generates Project Requirement Documents (PRDs) and the follow‑up Technical Specification (Tech Spec) through an agentic workflow. The goal is to compare vanilla LLM clients (OpenAI & Google GenAI) with popular agent frameworks (CrewAI, AutoGen, etc.) while visualising every step—outline, draft, critique, revision—in real time.
| Phase | Feature | Status |
|---|---|---|
| A | Vanilla workflow using official openai and google-generativeai Python clients |
🚧 In Progress |
| B | Pluggable adapter modules for frameworks (CrewAI, AutoGen, …) | ⏳ Planned |
| — | FastAPI backend that streams live updates via SSE | ✅ Implemented |
| — | Minimal front‑end (Streamlit) showing progress | 🚧 In Progress |
| — | One‑click “Generate Tech Spec” from the approved PRD | ⏳ Planned |
See docs/PRD.md for the full requirements.
- Docs – ✅ Finalise PRD and Tech Spec.
- Scaffolding – ✅ Set up repo structure, CI, and basic FastAPI + Streamlit apps.
- Phase A – 🚧 Implement the vanilla LLM workflow, including real LLM calls and connecting the frontend to the backend stream.
- Phase B – ⏳ Implement the first framework adapter (e.g., CrewAI).
- Tech Spec Generator – ⏳ Build the agent that converts the final PRD into a design doc.
- Evaluation – ⏳ Add more framework adapters & generate a comparison report.
Timeline details live inside the PRD.
backend/ # FastAPI app with routes, services, and agent pipeline
frontend/ # Streamlit UI application
agents/ # Shared agent protocols (interfaces)
docs/ # PRD.md, TECH_SPEC.md, and diagrams
.github/ # CI workflows and issue templates
We welcome issues and discussions.
A CONTRIBUTING.md guide will be added once the core features are stable.
MIT