Overview

The Server is the FastAPI Backend, served by Uvicorn and launched with mirumoji server (or uvicorn mirumoji.server.app:app)

It's Async-First → Routers are async def and blocking work is wrapped in asyncio.to_thread()

Layout

• App → FastAPI Application + Lifespan Handler

• Config → Runtime Configuration

• Constants → Shared Constants

• Dependencies → Dependency-Injection Helpers

• Media → Media File Handling

• Jobs → In-Process Async Job Queue + Worker That Backs The Single + Batch Operations

• Database → SQLAlchemy Models + Repos → Data-Access / Persistence Layer

• Models → Pydantic Schemas

  • JP Dict → Japanse Dictionary Data Models
  • Requests → Endpoint Requests
  • Responses → Endpoint Responses

• Processing → Transcription / Analysis Pipeline (Whisper, subtitles, text, LLM, audio, Anki) coordinated by the Processor

• Routers → HTTP Endpoints Grouped By Surface (audio, dictionary, health, jobs, LLM, profile, video)

• Modal Processing → Optional Modal GPU-Offload Entry Points + Configuration