# Architecture Overview

## Components

- `worlddisasterlm/` core ML package
- `backend/` FastAPI inference and risk APIs
- `frontend/` React disaster command dashboard
- `scripts/` operational entrypoints

## High-Level Flow

1. ETL collects and normalizes disaster records
2. Dataset builder creates instruction JSONL samples
3. Training pipeline fine-tunes base model using PEFT methods
4. Evaluation computes safety and quality metrics
5. Optimization exports ONNX/GGUF variants
6. API + dashboard deliver inference and analytics to users

## Design Goals

- Modular and scalable
- Enterprise-friendly and auditable
- Beginner-friendly with clear scripts and docs