digitaldemocracy2030/kouchou-ai/kouchou-ai-architecture

Kouchou-AI Architecture

Overview

Use this skill to orient yourself in the repo and explain how the system fits together. Remember that Kouchou-AI is a broadlistening system for the Digital Democracy 2030 project, adapted from Talk to the City for Japanese municipal use cases.

Core services and ports

• Locate the API (FastAPI) in apps/api/ and expect it on port 8000.
• Locate the public viewer (Next.js) in apps/public-viewer/ and expect it on port 3000.
• Locate the admin app (Next.js) in apps/admin/ and expect it on port 4000.
• Locate the static site builder in apps/static-site-builder/ and expect it on port 3200.
• Expect the optional Ollama service on port 11434 for local LLM usage.
• Expect Ollama to use the ELYZA-JP model by default.

Key directories

• Use packages/analysis-core/src/analysis_core/ for analysis steps, orchestration, services, and workflows.
• Use apps/api/broadlistening/pipeline/ for runtime configs, inputs, and outputs managed by the API.
• Use apps/public-viewer/components/charts/ and apps/public-viewer/components/report/ for report UI and charts.
• Use apps/admin/app/create/ and apps/admin/app/create/hooks/ for report creation UI.
• Use apps/api/src/routers/, apps/api/src/services/, apps/api/src/schemas/, and apps/api/src/repositories/ for API layers.

Pipeline architecture and report flow

• Follow the flow: CSV upload -> API validation -> pipeline run -> hierarchical output -> public viewer.
• Start API-side orchestration at apps/api/src/services/report_launcher.py.
• Start CLI-side orchestration at packages/analysis-core/src/analysis_core/__main__.py and packages/analysis-core/src/analysis_core/orchestrator.py.
• Expect API-managed outputs under apps/api/broadlistening/pipeline/outputs/{report_id}/.

Technology stack

• Treat the backend as FastAPI + OpenAI GPT models + sentence-transformers + pandas/numpy/scipy.
• Treat the frontend as Next.js 15 + TypeScript + Chakra UI + Plotly.js.
• Use pytest, Jest, and Playwright for testing.
• Use Biome (frontend) and Ruff (backend) for linting and formatting.
• Expect Azure Blob Storage support for storage needs.
• Use Lefthook for Git hooks (pre-push) as configured in lefthook.yml.

Configuration pointers

• Check biome.json, apps/api/pyproject.toml, lefthook.yml, and .env.example for core config.

Operational notes

• Require an OpenAI API key or local LLM for full pipeline runs.
• Validate LLM output for bias before acting on results.
• Back up report data before upgrading or applying breaking changes.
• Expect breaking changes between versions.